Catapulted through the tapestry slingshot network

Abstract

The present invention relates to network slingshot via a tapestry slingshot. A network system for providing long-distance network connection between endpoint devices is disclosed. The network system includes a first endpoint device, a second endpoint device, a first exchange server, a second exchange server, a first access point server coupled between the first endpoint device and the first exchange server, a second access point server coupled between the second endpoint device and the second exchange server, a first storage node coupled between the first exchange server and the second exchange server, and a second storage node coupled between the first exchange server and the second exchange server. The first exchange server is configured to convert a first packetized traffic into a carrier file and write the carrier file to the second storage node. The second exchange server is configured to read the carrier file from the second storage node and convert the carrier file into a second packetized traffic.

Description

[0001] Related Applications

[0002] This application claims priority to U.S. Provisional Application No. 62 / 327,907, filed April 26, 2016, which is incorporated herein by reference. This application claims priority to U.S. Provisional Application No. 62 / 327,846, filed April 26, 2016, which is incorporated herein by reference. This application claims priority to U.S. Provisional Application No. 62 / 327,911, filed April 26, 2016, which is incorporated herein by reference.

[0003] This application also relates to the following applications, the contents of which are incorporated herein by reference: International Patent Application No. PCT / IB16 / 01867 filed on December 9, 2016; International Patent Application No. PCT / US15 / 64242 filed on December 7, 2015; International Patent Application No. PCT / IB16 / 00110 filed on January 5, 2016; International Patent Application No. PCT / US16 / 15278 filed on January 28, 2016; International Patent Application No. PCT / IB16 / 00528 filed on April 7, 2016; International Patent Application No. PCT / IB16 / 00531 filed on April 7, 2016; International Patent Application No. PCT / US16 / 26489 filed on April 7, 2016; International Patent Application No. PCT / IB16 / 01161 filed on June 13, 2016. Technical Field

[0004] The present disclosure relates generally to networking and, more particularly, to a framework that utilizes the power of slingshot-propelled transmission into a transparent, optimized slingshot segment within a conventional Internet path. Background Art

[0005] One of the biggest advantages offered by current Internet technologies is that they adhere to standards that enable interoperability and communication between devices from different vendors. The underlying technology of Ethernet and the Internet Protocol (IP) based on it utilizes packets to be sent across segments between IP addresses as hops. These packets are vendor-independent, meaning they do not need to know or care who manufactured the gear they pass through, and can pass packets as long as the gear understands the structure of the packet and can parse its header and trailer.

[0006] Slingshot allows the benefits of Slingshot (see US Provisional Patent Nos. 62 / 266,060 and PCT / IB16 / 01867) to be used seamlessly for the long-haul portion of the transmission, while also allowing clients to use their existing equipment, configurations, and workflows.

[0007] Slingshot can be built independently on top of Slingshot, or it can be integrated with the Global Virtual Network (GVN) or some other network framework.When referring to a Slingshot Node (SLN), this can refer to a supercomputer cluster or node, a high performance computer or cluster, or other powerful computing system.

[0008] The two most common communication protocols on the Internet are the Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP) over IP, abbreviated as TCP / IP and UDP / IP, respectively. The TCP / IP protocol provides reliable round-trip transmission due to transmission control based on window size and round-trip time (RTT), built-in error correction, verification of the integrity of the transmitted data via an acknowledgment packet (ACK) returned to the sender, and other features; however, this reliability comes at a cost, as TCP / IP is significantly slower than UDP / IP. Although UDP / IP is relatively fast because it is a one-way transmission without an ACK sent back to the sender, there is no verification of data integrity, so the sender cannot detect loss or corruption at the receiver. Moreover, any packets lost via the UDP / IP route are completely unknown to the receiver. The main trade-off of TCP / IP over UDP / IP is accuracy versus raw speed.

[0009] While the most widely used protocols on the Internet are TCP / IP and UDP / IP over Ethernet, Ethernet itself has various versions, from the original Ethernet to GigE, 10GigE, 40GigE and 100GigE Ethernet are available, and of course higher bandwidth versions are expected in the future.

[0010] In other methods, hosts are "found" by other hosts through modern routing using IP addresses, and they can also be found via the Domain Name System which translates domain names into IP addresses. Routing protocols and route advertisements help devices determine IP path(s) between devices.

[0011] Over long distances, TCP / IP and UDP / IP protocols on Ethernet present certain challenges. Various technologies have been developed to try to force data to flow over the best path, including OSPF (Open Shortest Path First), BGP routing (Border Gate Protocol), and other peer-to-peer related technologies.

[0012] The Internet is an open network of multiple networks that peer with each other directly or via peering with third-party networks. There is little control over the flow of traffic through the networks of intermediary third parties. For those that require higher quality of service (QoS) and have strict performance requirements, dedicated lines and related technologies such as MPLS (Multi-Protocol Label Switching), dark fiber and other options provide direct connectivity between points, guaranteeing QoS, alleviating congestion from other traffic and having other guarantees, albeit at a relatively higher monetary cost (CAPEX and recurring operating costs) than regular Internet connections.

[0013] WAN optimization is another technology that some parties use to improve connections. WAN optimization equipment hardware and software are used to compress and optimize the data transmitted between two Internet edge points in a WAN so that they can push data across that link more efficiently.

[0014] The Global Virtual Network (GVN) (see International Patent Application No. PCT / US16 / 15278) comprehensively optimizes system-wide connections between connected devices, thereby significantly improving network performance by leveraging technologies such as artificial intelligence (AI) and advanced smart routing (ASR).

[0015] All of the above techniques follow current communication methods for transmitting data as packets between an origin and a destination, where the round trip and transmission time back to the origin is reflected as a measure of round trip time (RTT).

[0016] Long distance InfiniBand (IB) is another option, where two end-point InfiniBand enabled devices are positioned at both ends of a dark fiber channel to achieve a long distance IB connection. There are other protocols, such as Fiber Channel, but at this time it is unreliable over long distances.

[0017] Slingshot unidirectional transmission (see U.S. Provisional Patent No. US62 / 266,060 and International Patent Application No. PCT / IB16 / 01867) provides certain advantages for near-line-speed movement of data by sending the data to be written unidirectionally via remote direct memory access (RDMA) to a parallel file system (PFS), RAM drive, hierarchical file system (HFS), or other storage or queue. The written data is read by the slingshot node (SLN) in the remote area where it will be used. Slingshot can operate over an IB long-distance link or a high-speed Ethernet link that can support RDMA and other attributes, or over other protocols.

[0018] Catapult routing is an extensible layer built on top of Slingshot to allow for reasonable and clean routing to multiple points via the Catapult routing addresses. The present invention is built on Slingshot technology, but can also utilize GVN, Catapult routing, and other technologies.

[0019] Beacon pulsators are another technology built on top of Slingshot (see US Provisional Patent Application No. 62 / 266,060) and are used for unidirectional broadcast transmission of data.

[0020] As further background, the laws of physics serve as the basis for references to time, delay, line speed, and other time-related metrics herein. Because time and distance are important, the present invention uses the following baselines for time and distance / time basis: distances herein are measured in miles under the imperial system; distance measurements herein may be numbers with or without commas and / or decimals, or may be expressed as integers; one exception to distances not using the imperial system herein is in the refractive index of optical cables, where distances are expressed in meters under the metric system; time is measured in seconds, expressed as integers, fractions, and / or decimals of seconds (e.g., instantaneous granularity of time may be measured as fractions (every 1 / 20 or 1 / 10 or 1 / 100) or decimals (0.05, 0.1, 0.01) of milliseconds); time units cited herein may also be granularity finer than seconds, such as milliseconds (ms) and microseconds (μs), nanoseconds (ns), etc.; any granularity finer than microseconds, such as nanoseconds (ns), may be important in certain practical applications of the present invention, but for purposes of illustration, the finest practical granularity herein is μs.

[0021] In computing, the most common measure of time for networking is milliseconds (ms), while for processing it is microseconds (μs) or less.

[0022] The following table shows some possible values ​​and their corresponding equivalent conversions.

[0023] Table 1 - Measurement of time

[0024] # illustrate Second Milliseconds (ms) Microseconds (μs) 1 1 / 10 of a second 0.10000 100 100,000 2 1 / 20 of a second 0.05000 50 50,000 3 1 / 100 of a second 0.01000 10 10,000 4 10 microseconds 0.00001 0.010 10 5 100 microseconds 0.00010 0.100 100 6 1,000 microseconds 0.00100 1.000 1,000

[0025] At the heart of slingshot, pogo and related projectile techniques is the idea that synchronizing certain interdependent or sequential or other interval-based actions relies on actively focusing on momentary granularity (see U.S. Provisional Patent Application No. 62 / 296,257 and / or International Patent Application No. PCT / IB16 / 01867).

[0026] The inherent problems of the Internet Protocol (IP) are particularly evident over long distances for TCP / IP and UDP / IP protocols. In addition to distance-related problems, congestion through the equipment is another major cause of packet loss and other problems.

[0027] TCP / IP has a mechanism to resend lost or discarded or damaged packets. However, during times of congestion due to overuse, failures, or other reasons, the device that detects the loss will keep resending the lost packets, which can trigger a cascading effect that can exacerbate the problem, leading to more congestion.

[0028] Peering and weaving together many different segment characteristics along a network path can adversely affect the QoS of traffic flows throughout that path.

[0029] From a user experience (UX) and / or application performance perspective, high latency due to long paths, inefficient routing, congestion, loss, and other issues can cause service outages, application timeouts, and other current adverse effects.

[0030] Internet Protocol (IP) over Ethernet becomes extremely inefficient over long distances and its effectiveness decreases when there is congestion, poor routing, or other events.

[0031] The physical limitations of the lines also present challenges: Due to the laws of physics, light traveling over fiber-optic lines cannot reach the speed that light can reach in a vacuum.

[0032] Table 2 - Fiber line speeds taking into account the delay in fiber delay due to refraction (based on data found at Data: http: / / www.m2optics.com / blog / bid / 70587 / Calculating- Optical-Fiber- Latency )

[0033]

[0034] The table above compares the speed of light in a vacuum to the speed of light inside the glass core of an optical fiber. Therefore, there are physical limits on the efficiency of optical fibers, which establish a baseline for the theoretical best speed that light can achieve traveling through an optical fiber.

[0035] Although the refractive index of the fiber optic cable may vary slightly, the average value is assumed as follows: For an average efficiency of 68.05%, the average value is approximately 203m to 204m / μs, which is equivalent to the speed of light of 299.792m / μs.

[0036] Therefore, the transmission speed over optical fiber is 126,759.88 miles per second, which is in itself the fastest possible line speed that can be achieved.

[0037] On today's Internet, the request-response nature of host-to-host round-trip transmissions (and the corresponding RTT measurement of elapsed time) requires one host to query another host for information to be returned. As a result, host-to-host communications and, by extension, delays along the path lead to inefficiencies.

[0038] The packetization of data traffic also leads to inefficiencies. When the size of the message data (in bytes) exceeds the number of bytes allowed for the data payload of each packet, the packet size limit requires that the data message be split into multiple packets. In the IP model (see Figure 3 ) or the OSI model (see Figure 4 ) at each of the various levels of the , there is also significant bloat added by the packet headers (and possibly the file trailer), significantly affecting the payload size in bytes.

[0039] The Maximum Transmission Unit (MTU) specifies the largest number of 8-bit bytes that can be sent as a packet or frame. For example, if a packet has an MTU of 1,500, it means that the combined headers and trailers from all levels plus the payload cannot exceed that total. The minimum MTU is 68. And, the most common TCP MTU values ​​on the Internet today are 1500 or 576. Working backwards, the maximum allowed payload size in bytes can be calculated by taking the MTU and subtracting the combined header and trailer bytes, plus any additional bloat to determine the remaining bytes available for payload data bytes.

[0040] When the file size is larger than the allowed size of the data payload, then it is sent as a multi-part message. This requires breaking the file into fragments with more headers / footers added to indicate the beginning and end of each fragment of the file within the already constrained allowed size of the data payload. This bloat within bloat is severe when large files are to be sent over large distances.

[0041] TCP / IP and UDP / IP are subject to the Internet Protocol (IP) of the store-and-forward model of Ethernet, which processes packets through each device. This means that for each hop across the device, the packet must be received in its entirety, checked for integrity, and then forwarded forward. While this process only takes about 100μs per hop, the longer the path and the more hops the packet must pass through in each direction, the longer the RTT time. It is conceivable that a packet must be transmitted through 15 or 20 hops in each direction, thus increasing 2*15*100μs=3,000μs to 2*20*100μs=4,000μs. This means that the transmission time is increased by 3ms to 4ms due to the packet transmission equipment at the hop. In addition, if there is congestion, the device may take significantly longer to manage packet transmission.

[0042] Since boundaries are crossed between various network fabrics, routing, paths, peering issues and fabric related issues on many segments can cause problems. Private lines are point to point, and while most are optimal, they are closed and therefore do not suffer from the same congestion issues. In terms of the protocol, they are still request / response and therefore still subject to the above protocol limitations which are amplified over long distances.

[0043] While Slingshot is a new approach to solving many of the above problems, it is expensive for a single user to use on its own, both from a CAPEX perspective and from an ongoing operations perspective. In many cases, Slingshot requires a framework to be built on top of it. In some cases, such as data transfer utilization by financial companies, the expense and effort is worthwhile. However, many other industries cannot afford, nor do they need, a dedicated Slingshot mechanism.

[0044] Information beacon pulsators that use slingshots to transmit streaming or multicast data are good for one-way delivery of information, such as long-distance forwarding of marketing information. However, they are not suitable for integration into the Internet path where the recipient sends a response back. Summary of the invention

[0045] In accordance with the disclosed subject matter, systems and methods are provided for a framework that harnesses the power of slingshot propulsion transmissions into a transparent optimized slingshot segment within a conventional Internet pathway.

[0046] The bouncing in the present invention can transparently utilize the power of the slingshot as a segment in a pair of hops located in the path of a conventional IP hop. Bouncing is one or more long-distance segments that provide fast transmission over long distances relative to local Internet transmission. Bouncing is powered by a framework that consists of two slingshots, one slingshot sends data from one area to the destination area, and the other slingshot sends information back from the destination area to the original area.

[0047] At the entry point, the bounce captures and buffers the received packets into a queue. In a variable time period controlled by the tick granularity, the received packets accumulated in the queue buffer are combined into the file data part of a bulk uncompressed carrier file. The combined file is slingshotted to the other end of the bounce and saved in a folder on the PFS device where it is read by the Slingshot Node (SLN). Each time period controlled by the tick granularity on that SLN, the yet-to-be-combined carrier file is read from the PFS there and used as follows; each packet is personalized and returns on its journey by transmission to the next hop and is sent on the way back. As far as the next hop on the IP path is concerned, it is deemed to have received the packet from the equivalent IP device.

[0048] Routing between bounce ingress / egress IP address pairs is based on bouncing packets to PFS devices in the destination zone. Fig.12 , Fig.16 and Fig.17 The routing to the target area is based on which PFS device the carrier file containing the packet is sent to.

[0049] Bounce also provides full integration and interoperability with legacy systems running IP on the Internet, because the IP address pairing indicates the entry and exit points of the bounce. From the perspective of the IP device, the two addresses appear to be standard IP hops at either end of the IP segment, and the underlying bounce operation is transparent to them.

[0050] Slingshot nodes (SLN) are a feature of Slingshot. They can be supercomputer nodes (SCN), high performance computers / clusters (HPC) operating together, or other systems that can operate independently or as a group all at a lower level and jointly based on their relative situational roles from an overall functional perspective.

[0051] Slingshot is the Internet below (UTI squared or UTI 2 ) The second level. For a more thorough description see Figure 5 .

[0052] In some embodiments, the present invention takes into account the speed, power, and advantages of Slingshot over IP, but as a transparent relay embedded within the IP path.

[0053] In some embodiments, the invention features an algorithmic analysis that can determine whether data will be faster via a conventional IP path or if bouncing the transmission can provide a speed gain in the form of lower latency (see Fig.25 ).

[0054] In some embodiments, the present invention does not require any specific configuration of the sender or receiver of traffic passing through the transmission bounce transmission path. Therefore, achieving corresponding automatic speed advantages in the long-distance relay segments in the segment path is seamless.

[0055] In some embodiments, the present invention is a reasonably scalable product for sending extensive, high-volume data, achieving consistently high speed, high throughput because it operates at near line speeds.

[0056] In some embodiments, the present invention does not have the same congestion and other problems as IP, so bounces do not suffer the same inefficiencies over long distances that IP suffers.

[0057] In some embodiments, the present invention replaces most of the hops in the IP-based chain of segments, and instead it pops the data using simplified global routing over maximum distance.

[0058] In some embodiments, the present invention uses a slingshot "internal structure" that is protected by both the security perimeter and the nature of the protocol. Therefore, there is a security advantage because it is not possible to intercept, interfere with and / or modify traffic through the hopping segment (see Figure 2 ).

[0059] In some embodiments, the present invention proposes efficient bidirectional convergence of traffic between one bounce (egress ingress point) EIP and another bounce EIP.

[0060] In some embodiments, the present invention provides transparent segments in the IP path, and although packets are unaware of bounces in the middle, as a result it achieves the lowest possible RTT while also reducing the number of hops.

[0061] In some embodiments, the present invention results in better speed and performance because there is no long-distance efficiency reduction that typically occurs with long-distance IP transmission.

[0062] In some embodiments, the present invention is low-cost to users because the cost of the infrastructure is spread out and borne by many clients that share the infrastructure.

[0063] In some embodiments, the present invention alleviates the concerns of many organizations that have invested heavily in infrastructure built on Ethernet-based Internet Protocol (IP). Due to cost and the requirement for interoperability with legacy systems, they are reluctant to abandon their existing systems in favor of new systems, so the plug-in enhancements provided by Bounce over long distances allow them to use most of their legacy systems.

[0064] In some embodiments, the present invention is a bidirectional hybrid of traditional network segments boosted by (multiple) bouncing transparent segments, providing a boost to facilitate doing so.

[0065] In some embodiments, the present invention can be easily integrated into flexible, dynamic, secure and intelligent structures.

[0066] In some embodiments, a network system for providing long-distance network connection between endpoint devices is disclosed. The network system includes a first endpoint device, a second endpoint device, a first exchange server, a second exchange server, a first access point server coupled between the first endpoint device and the first exchange server, a second access point server coupled between the second endpoint device and the second exchange server, a first storage node coupled between the first exchange server and the second exchange server, and a second storage node coupled between the first exchange server and the second exchange server. The first exchange server is configured to convert a first packetized traffic into a carrier file and write the carrier file to the second storage node. The second exchange server is configured to read the carrier file from the second storage node and convert the carrier file into a second packetized traffic. The first exchange server is configured to (1) convert the first packetized traffic received from the first endpoint device into a carrier file via the first access point server, and (2) write the carrier file to the second storage node. The second exchange server is configured to (1) read the carrier file from the second storage node, and (2) convert the carrier file into a second packetized traffic, wherein the second packetized traffic is sent to the second end device via the second access point server.

[0067] There has thus been outlined, rather broadly, the features of the disclosed subject matter in order that the detailed description thereof that follows may be better understood, and in order that the present contribution to the art may be better appreciated.

[0068] In this regard, before explaining in detail at least one embodiment of the disclosed subject matter, it should be understood that the disclosed subject matter is not limited to its application to the details of construction and the arrangement of components set forth in the following description or shown in the accompanying drawings. The disclosed subject matter can have other embodiments and can be practiced and implemented in various ways. Moreover, it should be understood that the wording and terminology employed herein are for descriptive purposes and should not be considered limiting.

[0069] Thus, those skilled in the art will appreciate that the conception upon which this disclosure is based may readily be utilized as a basis for the designing of other structures, methods and systems for carrying out the several purposes of the disclosed subject matter.

[0070] For a better understanding of the disclosed subject matter, its operating advantages and specific objectives attained by its uses, reference should be made to the accompanying drawings and descriptive matter in which preferred embodiments of the disclosed subject matter are illustrated. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to facilitate a more complete understanding of the present disclosure, reference is now made to the accompanying drawings, in which like elements are referenced with like numerals or reference characters. These drawings should not be construed as limiting the present disclosure, but are intended to be illustrative only.

[0072] Figure 1The prior art of slingshots is shown;

[0073] Figure 2 A security perimeter with an upper GVN and a lower ejection infrastructure layer is shown;

[0074] Figure 3 shows the packet expansion in each layer of IP;

[0075] Figure 4 shows packet expansion in each layer of the Open Systems Interconnection (OSI) model;

[0076] Figure 5 shows the Below the Internet (UTI) bridging layer mapped to the Over the Top (OTT) layer;

[0077] Figure 6 Visualize the advantages of slingshot delivery via the internet by comparing it to a train at an airport versus a person walking a long distance to their gate;

[0078] Figure 7 Shows how the Slingshot is integrated into the IP path as a bounce;

[0079] Figure 8 The advantages of integrating bounce into long-range paths are shown;

[0080] Fig. 9 The bouncing integration is shown as a segment between two hops;

[0081] Fig.10 Three different network textures are shown;

[0082] Fig.11 The round trip path of the slingshot is shown in steps to describe each different operating part of the ejection;

[0083] Fig.12 The process of a bounce operation is shown, and its conversion of packets into files, and then conversion of files back into packets in both directions of the bounce;

[0084] Fig.13 Detailed by adding a timeline comparison between the Internet and the ejection Fig.12 ;

[0085] Fig.14 Several connection options available between two global nodes are shown;

[0086] Fig.15 It shows the backbone connections between various important financial markets;

[0087] Fig.16 The Internet paths and bounce paths between three global nodes are shown;

[0088] Fig.17 Shows how Global Virtual Networks (GVNs) can be combined with technologies such as Network Slingshot to achieve huge advantages seamlessly over long distances;

[0089] Fig.18 shows how various devices are addressable in each zone by their type and location;

[0090] Fig.19 The internal structure and operation of Bounce in terms of topology are shown;

[0091] Fig. 20 An example showing how to find a bouncing node;

[0092] Fig.21 A global virtual network (GVN) or similar global distributed network using a hub-and-spoke topology with octagonal routing on a backbone is shown, with egress / ingress points (EIPs) indicated;

[0093] Fig. 22 Describes the physical topology of the global node ring;

[0094] Fig.23 shows how packets of various sizes can be contained within a bounce carrier file;

[0095] Fig.24 Describes the reading process of the ejection node (SLN) during the ejection process;

[0096] Fig.25 describes in more detail the timing sequence of reading files by the ejection node;

[0097] Fig.26 The batches at variable intervals (not fixed times) are shown;

[0098] Fig. 27 Batch processing of files is shown, with the individual files shown being added;

[0099] Fig.28 An algorithm for evaluating which transport method is more efficient is described;

[0100] Fig.29 Describes the modules that run on various devices to power the integration of Bounce into GVN. DETAILED DESCRIPTION

[0101] In the following description, many specific details about the systems, methods, and media of the disclosed subject matter and the environments in which these systems, methods, and media can operate are set forth in order to provide a thorough understanding of the disclosed subject matter. However, it will be apparent to those skilled in the art that the disclosed subject matter can be practiced without these specific details, and certain features known in the art are not described in detail to avoid complications of the disclosed subject matter. In addition, it will be understood that the examples provided below are exemplary, and it is contemplated that there are other systems, methods, and media within the scope of the disclosed subject matter.

[0102] Figure 1 Two or more slingshot nodes working together to form a slingshot basis are shown. The figure describes the prior art of slingshot, where files of any size are directly written to remote PFS devices in the target area, 1-606 and 1-602, via one-way data transfer of RDMA (1-W606 or 1-W602) through write queues 1-WQ502 and 1-WQ506. Files are accessed on PFS1-606 or 1-602 through read queues 1-RQ506 and 1-RQ502, respectively. The figure shows the operation of two independent slingshot mechanisms (see U.S. Provisional Patent Application No. 62 / 266,060 and International Patent Application PCT / IB16 / 01867), which are juxtaposed and overlapped into an integrated relationship.

[0103] Traffic flows from GVN 1-322 to GVN 1-326 as follows. Traffic flows from GVN 1-322 to access point server (SRV_AP) 1-302 via 1-P322 and to backbone switch server (SRV_BBX) 1-502. At this point, the slingshot mechanism on SRV_BBX / slingshot node (SLN) 1-502 converts the packetized traffic into a combined carrier file via its write queue 1-WQ502 function and writes this carrier file directly to a folder on the parallel file system (PFS) storage node 1-606 via path 1-W606. The Read Queue 1-RQ-506 function of the SRV_BBX SLN 1-506 retrieves the carrier carrier file from the folder on the PFS 1-606, which then separates the carrier carrier file back into individual packets which are sent to the SRV_AP 1-306 via path 1-P506 and then to the GVN 1-326 via 1-P326.

[0104] Traffic in the other direction flows from GVN 1-326 to GVN 1-322 along the path described below. Traffic flows to the access point server (SRV_AP) 1-306 via 1-P326 and flows to the backbone switch server (SRV_BBX) / slingshot node (SLN) 1-506. At this point, the slingshot mechanism on the SRV_BBX SLN 1-506 converts the packetized traffic into a combined carrier file via its write queue 1-WQ506 function and writes this file directly to a folder on the parallel file system (PFS) storage node 1-602 via path 1-W602. The Read Queue 1-RQ-502 function of the SRV_BBX 1-502 retrieves the carrier file from the PFS 1-602, which it then separates back into individual packets, which are sent to the SRV AP 1-302 via path 1-P502 and then onto the GVN 1-322 via 1-P322.

[0105] QoS settings can be integrated to prioritize traffic by writing to a specific folder over another. The read queue can be configured to process files in one directory first, or it can also be set to check one folder more frequently than others, thereby enhancing the priority of its contents.

[0106] As described above, each unidirectional communication path is powered by a slingshot as defined in US Provisional Patent Application No. 62 / 266,060.

[0107] These two nodes and their corresponding communication paths working together form the basis of the underlying bouncing technology.

[0108] Figure 2 A security perimeter is shown with the GVN above and the Slingshot infrastructure layer below. The diagram shows the "air gap" between the Internet 2-822 and the backbone 2-832 over which the Slingshot protocol mechanism runs. Due to the differences in the protocols and the nature of how Slingshot operates, there is a security perimeter 2-182 between the two.

[0109] There is a security perimeter 2-182 located between the IP / Internet layer 2-822 and the BB / backbone layer 2-832. The security perimeter 2-182 can isolate the layers above from the layers below through firewall type operations. Another built-in protection involves the nature of the transmission. Packets travel along path 2-TR6AP, and files are written to the PFS device via RDMA via path 2-TR6BP. Files cannot be moved locally at the IP layer, and packets cannot be routed separately via the BB layer.

[0110] Figure 3Describes the packet expansion of IP transport packets when headers are added to the data at various layers. At the application layer 3-L04, the data payload has an initial size as indicated by data 3-D4. The size of the packet is indicated by the packet size parameter 3-PBytes.

[0111] At the next layer, transport layer 3-L03, the packet size 3-PBytes has the original size of the data 3-D4, which is equal to the data UDP 3-D3. It also includes the expansion of the header UDP 3-H3.

[0112] At the next Internet layer 3-L02, the main payload data IP 3-D2 is a combination of 3-D3 and 3-H3. It is increased by 3-PBytes through the header IP 3-H2.

[0113] At the link layer 3-L01, the frame data 3-D1 is a combination of 3-H2 and 3-D2, and the size of the frame data 3-D1 is further increased by 3-PBytes through the header frame 3-H1 and the file end frame 3-F1.

[0114] Figure 4 The packet expansion of headers and data for each of the seven layers of the OSI model is described. The original data 4-D0 grows with the application of OSI layer 7 4-L7 at each level by adding headers such as header 4-H7. In each subsequent layer from layer 7 down to layer 1, the data layer is a combination of the data and headers of the combined previous upper layer.

[0115] The total packet expansion in the OSI model at the physical OSI layer 14-L1 is indicated by the packet size 4-PBytes.

[0116] Figure 5 Shown is the Below Internet (UTI) Bounce Bridging Layer mapped to the Over-the-Top (OTT) layer. This diagram visualizes how GVN operates as the first level on top of the Internet and Slingshot operates as the first level below the Internet. Bounce is the second level below the Internet and is integrated as a segment in the GVN path or the regular Internet path. Figure 5 Indicates the level at which the Slingshot BB5-86 fits into the topology hierarchy. OTT 1 Indicates the top level of the Internet. OTT 2 Indicates the second level at the top of the Internet, which means it is OTT 1 The top of the component. UTI 1 Indicates the first level of the Internet layer. OTI 2 Indicated in OTI 1 The second level of the layer below the Internet below the component.

[0117] OTT and UTI are used for descriptive purposes only to indicate the layering of relationships and interactions. At the physical layer, all types of protocols can exist at the same level or at different levels. GVN 5-82 is OTT 1 5-TOP82 layer global virtual internet built on the basic pipeline of the underlying internet 5-TOP80 that powers ISP network connections 5-80.

[0118] The bouncing BB 5-88 mechanism is in UTI 2 The second level of the UTI of the 5-UNDER88 layer. It utilizes the UTI of the slingshot BB 5-86 1 Technology. Its functionality and corresponding advantages can be integrated into the business of GVN 5-82, which is in the OTT 1 5-TOP82 layer, or integrated as a segment in the Internet path at the base Internet 5-TOP80 level. Therefore, Slingshot is a hybrid bridge between GVN and Slingshot.

[0119] In OTT 2 The example of the OTT second level of the MPFWM 5-84 at the 5-TOP84 layer is indicated for illustrative purposes only. In a live implementation, it may or may not be integrated into the traffic flow.

[0120] Figure 6 The advantage of slingshot transport via the internet is visualized by comparing it to taking a train at an airport versus a person walking a long distance to their boarding gate. Due to the distance and the complex route, a walker can easily tire and get confused or lost, and it takes an indeterminate amount of time to get from starting 6-010 or 6-910 to ending 6-090 or 6-990. Those who choose to take the train have a short walk to the station, and if they choose to take the train at 6-P200 or 6-P700, a certain amount of boarding and alighting time will be added at 6-200 and 6-300 or 6-700 and 6-800, but the journey on the train at 6-P300 and 6-P800 is faster, more direct, less likely to get lost, and its travel time is deterministic than walking.

[0121] In particular, Figure 6An analogy as a visual graphic is provided, comparing the path that a passenger might take if they walk through the airport from the check-in 6-000 to the gate area 6-900 or the path that they might take if they take the train. Both start at 6-010 and can travel along the relay segment on the path from point DA to DF, and can go directly to 6-100 or can stroll. They can decide to take the train at 6-100. If they do this, they get on the train at 6-200 via the path 6-P200, which takes them directly to 6-300 via the path 6-P300, where they get off via the path 6-P320. From there they enter the walking path again at DW and go to the boarding area at 6-090 via other path relay segments DX, DY, DZ. Those who choose to walk via route 6-090 will begin walking at DG and will likely stroll as they walk between the various other relay points in their route until they also reach end 6-090. While getting on and off the train may add a little extra time, this is more than compensated for in the overall time due to the rapid nature of train transportation. Those who take the train will also experience less fatigue and potential stress than those who walk.

[0122] The path from the airplane 6-900 to the terminal exit 6-000 begins at start 6-910 and again offers the choice of taking the train or walking, with similar performance and time advantages for those who choose to take the train. This is an analogy for deciding whether to use bouncing between long distance points or to have packets travel along an extended Internet path.

[0123] Figure 7 Shows how Slingshot is integrated into an IP path as a bounce. It compares the Internet cross-region 7-CPT340 timing of 7-TWms to the bounce timing of 7-TV+7-TZ+7TV for processing at both ends of 7-CPT110 and 7-CTP-210 and the bounce 7-CPT300. The bounce is powered by the Slingshot mechanism as shown in this article, but as far as the client is concerned, the relay segment starts at 7-P502 and ends at 7-P506, as well as the return of 7-P502 and 7-P506. Segment 7-CPT306 replaces P0-4 to P0-10, while 7-CPT302 replaces P0-10 to P0-4.

[0124] Figure 7 Based on Figure 1 The depicted display of the hopping operation is constructed with the exception that the access point server (SRV_AP) is not identified, nor are any other devices identified above the backbone switching server (SRV_BBX) / bouncing node (SLN) 7-502 and 7-506.

[0125] Paths 7-P502 and 7-P506 link the bouncing mechanism to the end hop bridgeheads of segments P0-3 and P0-11, respectively. The two paths 7-CPT302 and 7-CPT306 between those end hops indicate Figure 1 Each one-way slingshot described. When combined into hops, these paths constitute segments, which appear to the client data as just another segment in the chain.

[0126] exist Figure 7 , there is a non-bounce enabled path between the entry / exit point on the left side of segment P-01 and another entry / exit point on the channel on the right side of segment P0-14.

[0127] Between the non-bounce enabled path and the bounce enabled path are bracket marks that denote the segment portions as blocks to compare the operation of the bounce enabled path to a conventional non-bounce Internet path.

[0128] The initial set of segments at one end of each path is the Internet local 7-CPT 140, which is compared to the path of SRV_BBX / SLN 7-CPT 100, and for this example, each has the same Δt X ms. The initial set of segments at the other end of each path is the Internet local 7-CPT 240, which is compared to the path of SRV_BBX / SLN 7-CPT 200. For this example, each has the same duration Δt Y ms.

[0129] The middle segment is where comparisons can be made between the bounce enabled path and the regular internet path. The regular internet path duration of the segment internet spanning region 7-CPT340 is Δt W ms. The bounce enabled segment is the combined sum of three durations: the processing on the SRV_BBX / SLN 7-CPT110 with Δt = +<1ms plus the slingshot hop segment 7-CPT300 with Δt = Zms plus the processing on the SRV_BBX / SLN 7-CPT210 with duration Δt = +<1ms. The rationale for adding the three durations is to not only take into account the time and efficiency gained by bounce on the slingshot hop segment 7-CPT300, but also the drag created by the bounce processing time at both ends. Even though more time is added for processing packets combined into files, which are saved to the remote PFS, and then pulling the files and breaking them back into packets, the key point is that the efficiency and speed gains from bounce due to its relatively short end-to-end duration in both directions will more than compensate for this.

[0130] Figure 8 Continuing to demonstrate the advantages of integrating bounce into long paths, Segment 8 - CPT300 shows how bounce is a transparent and bidirectional segment. Figure 8 based on Figure 7, many of the same components are labeled. The main difference between these figures is that Figure 8 Hidden in the figure is a bouncing mechanism connected by paths 8-CPT302 and 8-CPT306.

[0131] The new channel line for the end-to-end segment is at the bottom of the figure. Unlike the two separate arrows 8-CPT302 and 8-CPT306 on the center line, the bounce segment is represented by 8-CPT300.

[0132] When the bounce-enabled path is compared to long-distance transmission over conventional Internet, the speed benefit from the transparent "middle" bounce makes up for the low increase time at both ends of the bounce.

[0133] Fig. 9 A visual comparison is shown between relays in two regions connected by an Internet long-distance network segment 9-2030 or those in the same region connected by a hop segment 9-2838. Even though the underlying transmission over the long-distance P2038 is via hop powered by a slingshot, device B 9-028 finds it by the IP address of device C 9-038. Device C 9-038 finds it by the address of device B 9-028 via path P2830.

[0134] Path P2030 represents many hops along the Internet over a long distance - the diagram is not drawn to scale. And paths P2038 and P2830 together represent bidirectional slingshot segments that together constitute hops over a long distance.

[0135] Over long distances, Internet segment 9-2838 performs significantly better and faster than Internet long-distance network segment 9-2030.

[0136] Fig.10 Three different network fabrics are shown: GVN tunnel TUN 10-222 between SRV_AP 10-200 and SRV_AP 10-202; Internet paths 10-P220 to 10-P236 on the open Internet; slingshot segment 10-SL518 between slingshot nodes SLN 10-508 and SLN 10-518, and backhaul 10-SL508 between SLN 10-518 and SLN 10-508. These three fabrics are combined into a network tapestry in the GVN.

[0137] Tunnel TUN 10-222 is over the top (OTT) of the Internet and bounces using a reverse slingshot mechanism on the fiber backbone.

[0138] Algorithmic analysis can be applied to select which type of transmission is optimal over which path. This segment and path analysis can also be used to adjust various settings for optimal performance.

[0139] Fig.11 based on Figure 7 , and shows some of the steps of a round trip of a packet through two reverse slingshots that constitute a bounce, where the bounce is inserted into the regular Internet path. SRV_BBX can contain supercomputer node (SCN) functionality, or be connected to an SCN. A parallel file system (PFS) represents storage devices available in a global cluster that can be reached via remote direct memory access (RDMA) for reading and writing files using the slingshot technology as described in U.S. Provisional Patent Application No. 62 / 266,060 International Patent Application No. PCT7IB 16 / 01867.

[0140] Following the straight path in the figure, the IP packet naturally travels along the hops from P0-1 to P0-14 and in both directions.

[0141] A set of sequential steps are described here for traveling in one direction from 11-A to 11-G:

[0142] Step 11-A: Start of the path. Traffic travels through three segments P0-1, P0-2, and P0-3.

[0143] Step 11-B: Traffic enters the bounce at this point 11-B; from the client's perspective, the traffic will flow through segment 11-CPT306 to the relay segment at 11-F.

[0144] Step 11-C: It will be apparent to the client that traffic will be diverted from the open Internet channel at 11-B and sent to 11-C where the IP packets are captured at ingress and packaged within files for processing by the write queue 11-WQ502 of the backbone switch server 11-502. The traffic is encapsulated in a transport file (see Fig.23 ) and in step 11-D is saved to the parallel file storage (PFS) device 11-606.

[0145] Step 11-D: Read the complete file in batches according to Fig.24 , batches of files stored on SRV_BBX / SLN 11-502 via write queues (direct writes via RDMA are also possible) and complete files over a period of time (see tick granularity in U.S. Provisional Patent Application No. 62 / 296,257) are read from PFS 1-606 by SRV_BBX / SLN 1-506. See Fig.25, for batch reading of files and use 25-RQP00 via 25-RQP606. Once used, the files are marked via 25-Q606 by post-processing P 25-Q00. This marking of files is essential so that files are not read twice by two overlapping batches. For a batch during duration 25-T08 of WS ms. During the next batch duration 25-T18, the next file 25-RQP10 is read and used via 25-RQP616, marked after use by post-processing P 25-Q10 via 25-Q616.

[0146] Step 11-E: The file is retrieved by the Read Queue 11-RQ506 Manager on the Backbone Exchange Server (SRV_BBX) 11-506 and the file is divided into individual packets. The packets re-enter the Internet channel at 11-F from 11-E via path 11-P506. Thus, they are re-packetized and sent on their way.

[0147] Step 11-F: Each packet(s) re-enter and then travel through standard Internet channels from IP addresses 11-F to 11-G via segments P0-11, P0-12, P0-13 to P0-14.

[0148] Step 11-G: Each packet travels through 11-G to the destination via the path of SRV_BBX / SLN 11-506 in the Internet local 11-CTP 140 via the path 11-CPT 200.

[0149] The following sequential steps for traveling in the other direction from 11 -H to 11 -N are described here.

[0150] Step 11-H: Return packets or packets near the 11-H trip from the Internet travel via the path P0-14 to P0-13 to P0-12 to P0-11 and start bouncing at 11-I.

[0151] Step 11-I: Packets enter the bounce via the IP address at end 11-I of segment P0-11 and are passed to SRV_BBX / SLN 11-506 where they are assembled into the payload of a "carrier" file consisting of various packets.

[0152] Step 11-J: The combined "carrier" file of the various packets is saved from SRV_BBX / SLN 11-506 to PFS 11-602 via write queue 11-WQ506 via path 11-W602.

[0153] Step 11-K: Write queue 11-RW502 periodically extracts batch files from PFS 11-602. In step 11-K, the operation on SRV_BBX 11-502 is very similar to 11-D.

[0154] Step 11-L: The "Carrier" file is retrieved by the Read Queue 11-RQ502 Manager on SRV_BBX / SLN 11-502 and is divided into individual packets. The packets re-enter the Internet channel at 11-M from 11-F via path 11-P502. Thus, they are re-packetized and sent on their way.

[0155] Step 11-M: Each packet reenters and then travels through the standard Internet channel from the IP addresses 11-M to 11-N via segments P0-3, P0-2 to P0-1.

[0156] Step 11-N: Each packet travels through the path of SRV_BBX / SLN 11-502 in the Internet local 11-CTP 140 through 11-N to the destination via the path 11-CPT 100.

[0157] The indicated time comparisons and other elements are based on Figure 6 , and are included in this diagram only as a reference point. The key points here are the steps listed above.

[0158] From a packet perspective, although 11-CPT306 and 11-CPT302 replace many intermediate hops and benefit from the efficiency of the slingshot technology, they are each just another hop between two hops.

[0159] Fig.12 yes Fig.11 3-1 to 3-8 are received by the slingshot and combined into a RESP batch 12-332 to be separated back into individual response packets P4-1 to P4-8.

[0160] As an illustration, 11-CPT 306 is a bounce of a request from an origin before segment P0-1 to a target other than P0-14, where a response of 11-CPT 302 is returned via the bounce.

[0161] This traffic corresponds to the path of a series of packets received via the SRV_BBX / SLN via paths P1-1 to P-1-8 and their subsequent buffering into a combined file via the write queue. The file composed of REQ1 to REQ8 packets is sent in one direction via storage to PFS 11-606 by bouncing REQ batches 11-312.

[0162] This traffic corresponds to the path of a series of packets received via the SRV_BBX / SLN via paths P1-1 to P-1-8 and their subsequent buffering into a combined file via the write queue. The file composed of REQ1 to REQ8 packets is sent in one direction via storage to PFS 11-606 by bouncing REQ batches 11-312.

[0163] Starting from the exit point of REQ batch 12-312, the personalized packets follow paths P2-1 to P2-8 to their corresponding destination addresses on the Internet 12-322. Assuming that the target hosts are all active and responding, their return packets are located and re-enter the bounce transparent segment via paths P3-1 to P3-8 to be reassembled into a file, which is then transmitted across path RESP batch 12-332 via file storage. Once at the other end of transparent bounce, the file is read and the packets are each returned to their originating device via paths P4-1 to P4-8.

[0164] Fig.13 based on Fig.12 , where a relative time comparison is added to evaluate the utility of bouncing transparent hops (processing on SRV_BBX 13-CPT10 plus bouncing hop 13-CPT300 plus processing on SRV_BBX 13-CPT210) relative to transmission over the Internet cross-region 13-CPT340 segment. In this figure, references to SRV_BBX can refer to a standalone SRV_BBX, a bouncing node (SLN), or a combination of both.

[0165] In some embodiments, the packets are combined, encapsulated, or otherwise concatenated into a carrier file containing the slingshot payload, and then the slingshot file is transmitted, while on the other side, the contents of the payload are separated back into their respective parts and then the packets are forwarded.

[0166] The elements 13-TW, 13-TX, 13-TW, 13-TY, and 13-TZ in the figure are directly related to the elements TV, TX, TW, TY, and TZ in the following equations, Equation 13-1, Equation 13-2, and Equation 13-3.

[0167] The total path delay of the Internet path can be derived from the sum of the durations Δt of the Internet local 13-CPT140 plus the Internet cross-regional 13-CPT340 plus the Internet local 13-CPT240.

[0168] Formula 13-1:

[0169] Total Internet path time Δt(Ρ004) = Δt(ΤX ms) + Δt(TW ms) + Δt(TY ms)

[0170] The total path delay of the bounce-enhanced hybrid Internet + bounce path can be calculated by adding the path of SRV_BBX 13-CPT100 plus the processing on SRV_BBX 13-CPT110 plus the sum of the slingshot hop 13-CPT300 plus the processing on SRV_BBX 13-CPT210 plus SRV_BBX to the Internet 13-CPT200.

[0171] Formula 13-2:

[0172] Total Internet-Bounce Hybrid Path Time (P504) = Δt(TX ms) + 2*Δt(TV ms) + Δt(TW ms) + Δt(TY ms)

[0173] Δt 13-TX ms is equivalent for the paths of the regional internet local 13-CPT140 and the SRV_BBX 13-CPT100.

[0174] Δt 13-TY ms is equivalent for the paths of Regional Internet Local 13-CPT240 and SRV_BBX 13-CPT200. Processing steps 13-CPT110 and 13-CPT210 indicate a short-term RAM cache for the group.

[0175] From the client's perspective, Slingshot hop 13-CPT300 is a transparent hop.

[0176] Fig.14Some of the connection options available between two global nodes 14-GN02 and 14-GN04 are shown. The figure visualizes the shorter time it takes to write to PFS 14-604 from a region 14-GN02 versus the time it takes for an Internet packet to be transmitted via 14-P004, and from 14-GN04 back to PFS 14-602. One connection option is via the Internet through path 14-P004, and another connection option is via hops through path 14-P504. Slingshot path 14-P504 indicates the latency between two backbone nodes 14-BBN02 and 14-BBN04 via hops powered by a slingshot or equivalent connection on a backbone fiber. Internet path 14-P004 represents connectivity via multiple different path options through the Internet and their corresponding latency ranges between those points.

[0177] This exemplary embodiment shows how Fig.13 When comparing the total delay of path 14-P004 to path 14-P504, using Fig.13 For simplicity, the two processing steps of the processing on SRV_BBX (and SLN) 13 -CPT110 and the processing on SRV_BBX (and SLN) 13 -CPT210 consume the same duration.

[0178] Therefore, the bounce improvement at the slingshot hop 13-CPT300 requires the time of the processing steps including the processing at the SRV_BBX 13-CPT110 and the processing at the SRV_BBX 13-CPT210, thus requiring 13-TV*2.

[0179] based on Fig.13 , the duration (Δt) of the Internet cross-region 13-CPT340 segment is labeled 13-TW and is measured in milliseconds (ms). In the figure, this is labeled Δt{TW ms}. The duration (Δt) of the slingshot relay segment 13-CPT300 is labeled 13-TZ in the figure, as well as its related processing segments "Processing on SRV_BBX" 13-CPT1 10, 13-CPT210. These are also measured in ms. In the figure, this is labeled Δt{TV+TZ+TV}.

[0180] Therefore, for bouncing to be effective, it must be more efficient and have lower latency than connections over the open internet:

[0181] Formula 14-1:

[0182] Δt{TV+TZ+TV}<Δt{TW ms}. What this equation means is the combined time for: TV, which is the process of adding items to and preparing the Slingshot payload; TZ, which is the sending of the payload through the Slingshot; and TV, which is the processing of the payload to separate back into the separate elements. The sum of these three parts is the total time for Slingshot. TW represents the time it takes for a packet to travel through the Internet via an IP path. TV at both ends adds time. TZ is fast transmission. The point is that the marginal time added by TV at both ends should be compensated by the faster time of TZ. When comparing TV+TZ+TV, the net total should still be less than TW

[0183] Therefore, Δt{TV+TZ+TV} must be less than Δt{TW ms}. The basic principle is that the increased duration added by the necessary processing Δt(TV) at both ends of the bounce must be offset by the speed advantage provided by the slingshot part of the segment Δt(ΤZ) between the bridgeheads at the end points of the bounce.

[0184] There are other advantages provided by bouncing, such as wide parallel bandwidth, which is more efficient than packetizing IP traffic over long distances, but this figure does not depict.

[0185] Fig.15 Demonstrates backbone connections between various important financial markets. The diagram shows the geographic locations and exemplary connection paths of several global nodes of the GVN. For illustration purposes, lines are drawn as straight lines between points.

[0186] In reality, the actual route of a pipeline is rarely straight or direct due to political / administrative boundaries, city limits, zoning, geographic features (such as bodies of water), various elevation changes, and other reasons. However, in most cases the additional distance caused by the path deviating from the potential most direct route does not add enough distance to have a significant adverse effect of additional delay.

[0187] Segments can be described as city / location pairs, and for bounce purposes, the origin endpoint of the bounce is represented by the IP address of a server or gateway device there, where the segment is transmitted via the bounce segment to the IP address of a server or gateway device in the target endpoint city / location.

[0188] Transmission from one location to another is as simple as going from the origin IP address to the destination IP address, and for the return path, the IP addresses are in reverse order. This single segment replaces many other IP segments on the Internet and is optimized by bouncing.

[0189] PFS names can be based on the last octet or the last two octets of the IP address. PFS names can also include city codes, regions, IP addresses, identified world nodes, and more. IP address pairs identify the bridgeheads at both ends of the segment.

[0190] For example, from 188.xxx.xxx.128 to 188.xxx.xxx.100 is by writing to PFS15-600. And if going back, from 188.xxx.xxx.100 to 188.xxx.xxx.128 is by writing to PFS15-628.

[0191] Like a round-trip route, the combination of two unidirectional segments constitutes a bounce-transparent round-trip integration nested within an existing IP tunnel.

[0192] Fig.16 The Internet channels and hop paths between the three global nodes 16-GN02, 16-GN04, and 16-GN06 are shown. Fig.14 , and added additional regional locations. All three PFS devices are available for Slingshot. It compares the delivery efficiency and speed advantages between Slingshot and the Internet. Paths 16-P004 and 16-P006 are via the Internet. Paths 16-P504 and 16-P506 are via Slingshot.

[0193] Fig.17 It shows how a Global Virtual Network (GVN) can be combined with technologies such as Network Slingshot to achieve great advantages seamlessly over long distances. Network Slingshot is further described in US Provisional Patent No. US62 / 266,060 or International Patent Application No. PCT / IB16 / 01867.

[0194] Fig.17 It describes that the global virtual network GVN 17-028 connects the LAN 17-376 via the endpoint device EPD 17-378 or the outgoing / incoming path with the SRV_AP 17-368 via the EIP 17-322. It also describes how the SRV_AP 17-302 is the front "head" to connect to the catapult powered slingshot and or bounce. The backbone switching server SRV_BBX 17-502 is connected to the catapult node SLN 17-538 and the global catapult backbone 17-P506 is connected to 17-P504 and the PFS device 17-602.

[0195] The first boundary is the GVN egress entry point (EIP) 17-322 between the Internet 17-328 and the GVN 17-028 or the EPD 17-378 between the LAN 17-376 and the GVN 17-028. The next boundary is the security perimeter 17-182. This layered security approach protects the core infrastructure on which the GVN is based.

[0196] The secure perimeter 17-182 boundary between the GVN and the GVN backbone protects the high-speed global network. The portion of the GVN above the perimeter 17-822 has traffic flowing over the top of the open Internet (OTT) via secure GVN tunnels. Below the secure perimeter 17-182, the GVN connections utilize various protocols over dark fiber or other connections that are not directly available from the Internet.

[0197] The launch node (SLN) 17-538 can be a supercomputer / cluster node (SCN), high performance computer / cluster (HPC), or equivalent, which operates inside (below) the security perimeter 17-832, which can operate a true internal network with advanced capabilities (such as remote direct memory access (RDMA)) to a parallel file system (PFS) 17-602 device.

[0198] The global ring of PFS storage devices passes through 14-P506 and 14-P504 and other nodes not mentioned here.

[0199] The SRV_AP 17-302 feeds information to and / or retrieves information from the SRV_INFO 17-338.

[0200] The security perimeter 17-182 sits between the IP traffic on the Ethernet network above the SP 17-182 and the non-IP internal slingshot / bounce traffic below it.

[0201] Fig.18 yes Fig.16 , which shows a topology of the bouncing mechanism indicating the placement of PFS devices 18-602, 18-604, and 18-606. It further shows how they can be connected directly through one segment 18-P504 or via two or more connecting segments 18-P504 to 18-P506. The key point is that a bouncing node in region 18-GN02 can write to PFS 18-604 in region 18-GN04 or PFS 18-606 in region 18-GN06, etc. These writes are direct and do not have to be relayed between PFS devices in intermediate regions.

[0202] Fig.19 The internal structure and operation of the bounce with respect to the topology are shown. Fig.19Not to scale. It shows how the Backbone Switch Server (SRV_BBX) and Sling Nodes (SLN) 19-502 to 19-516 can access and write to various PFS devices, such as 19-602 to 19-616. They are all connected via the internal backbone of various connection segments 19-P502 to 19-P516. As an example, it shows how Sling is integrated with the GVN and some of its devices, such as the Access Point Server (SRV_AP) 19-302, the End Point Device (EPD) 100, and the Central Control Server (SRV_CNTRL) 200.

[0203] Fig. 20 based on Fig.19 , with some exceptions. The key point is that the bounce node can be found by, for example, IP address XXX02 on BBX / SLN 20-502 or IP address XXX10 on SRV_BBX / SLN 20-510. The difference between these exemplary embodiments is that most of the bridgehead node points are faded to show availability, with two bridgehead nodes indicating a bounce connection from zone 2 20-ZN02 to zone 10 20-ZN10 via SRV_BBX / SLN 20-502 to write directly to PFS 20-610 via RDMA, where SLN / SRV_BBX 20-510 reads the carrier file and uses it in zone 10 20-ZN10. Reverse traffic in the other direction from zone 10 20-ZN10 to zone 2 20-ZN02 is written to PFS 20-602 via RDMA through SRV_BBX / SLN 20-510. The carrier file is read by SRV_BBX / SLN20-502 for use there.

[0204] These bridgeheads are bolded to highlight their location and focus. For illustration purposes, IP address XXX02 is indicated on SRV_BBX / SLN 20-502 and XXX10 is indicated on SRV_BBX / SLN 20-510 as both ends of the bounce. Therefore, the bounce from area 2 to area 10 follows the IP sequence of XXX02 to XXX10, while the bounce from area 10 to area 2 follows the IP sequence of XXX10 to XXX02.

[0205] Slingshot matches the target area and maps it to the PFS drive 20-610 or 20-602, which will be accessed through the read queue on the SLN on the remote side.

[0206] Fig.21A global virtual network (GVN) or similar globally distributed network is shown using a hub and spoke topology with octagonal routing on the backbone with egress / entry points (EIPs) indicated. The octagon shape is used for illustration purposes only to Fig. 20 Associated with a wider image.

[0207] Fig.21 The network topology of the GVNs in two different regions 21-RGN-A and 21-RGN-B and how these regions are connected by the global connection 21-RGN-ALL via paths 21-P0A and 21-P0B are shown. Fig.21 The hub & spoke connections in each of the two regions are shown. The multiple Exit-Entry Points (EIPs) 21-EIP400, 21-EIP420 and 21-EIP410, 21-EIP430 in each region are added spokes to the hub and spoke model.

[0208] SRV_BBX 21-280 and SRV_BBX 21-282 are backbone switching servers and provide global connectivity. SRV_BBX can be placed in a region serving as a global link as one or more load balancing servers.

[0209] The access point servers (SRV_AP) 21-302, 21-304 and 21-306 in 21-RGN-A are connected to SRV_BBX 21-280. The central control server (SRV_CNTRL) 21-200 serves all devices in the area, and it can be one or more multi-master SRV_CNTRL servers. The endpoint devices (EPD) 21-100 to 21-110 will connect to one or more multi-SRV_AP servers through one or more multi-concurrent tunnels.

[0210] The diagram further shows multiple egress entry points (EIPs) 21-EIP420, 21-EIP400, 21-EIP430, and 21-EIP410 as added spokes to the hub and spoke model, with paths in and out of the open Internet. This topology can provide EPD connections to EIPs in remote regions routed through the GVN. In the alternative, the topology also supports EPD connections to EIPs in the same region, EPDs in the same region, or EPDs in remote regions. These connections are securely optimized through the GVN. This also facilitates reaching the EPD from the open Internet, where traffic enters the EIP closest to the source and is carried via the GVN, realizing the benefits of GVN optimization.

[0211] exist Fig.21In RGN, RGN represents ring global node(s) or regional global node(s), and RGN_ALL represents all linked global nodes. Global nodes can be managed by MRGN, which is the manager of regional global nodes or a mesh of regional global nodes.

[0212] Fig. 22 Describes the physical topology of the global node ring. This diagram simplifies Fig.19 and Fig. 20 , showing the key points of the global backbone 22-P502 to 22-P516, PFS drivers 22-602 to 22-616 accessed by SRV_BBX / SLN 22-502 to 22-516. Each global node consists of an SRV_BBX device and its associated PFS device. Global nodes are connected to each other directly or via one or more of the other global nodes (multiple).

[0213] exist Fig.21 In , RGN is the "regional global node", and element 21-RGN-ALL stands for "all linked regional global nodes". Fig.21 The octagonal ring represented at element 21-RGN-ALL is enlarged to provide a more detailed view.

[0214] Routing is based on writes to the destination PFS device via RDMA, and subsequent file reads through the SRV_BBX / SLN in the target region and subsequent use of the data there.

[0215] For illustration purposes, the global structure of a PFS connection is an octagon, but in reality, any shape may be used. The bouncing mechanism depends on how the various devices are connected at the physical layer.

[0216] based on Figures 19 to 21 , the octagonal ring represents the routing options for bouncing.

[0217] Fig.23 Describes how packets of various sizes can be contained within a bounce carrier file (eg, various Internet structures can be connected to a bounce node). Fig.23A "carrier" file sent via Slingshot is described, consisting of a payload grouped in body data 23-200. This exemplary embodiment describes a data file organized in three defined parts: header information 23-100, a payload containing the body data 23-200, and a trailer 23-300. The file may be stored in RAM, other memory, saved to disk, or otherwise stored in memory or storage. The header may contain information about the host source, host destination, timestamp, and other information. Security information may be stored in fields in the header and trailer portions. The security information may hold notes on keys used for decryption, as well as other information. The payload (body data) may be encrypted in whole or in part, or may be sent unencrypted. The payload checksum in the trailer is used to verify the integrity of the body data. An EOF symbol in the trailer will indicate that the file has arrived, is complete, and is ready for verification / verification of accuracy before being ultimately used.

[0218] Fig.23 Various packets are shown, small packet sizes such as packets 23-A, 23-C, 23-E and 23-F, medium packet 23-D or larger packets such as packet 23-B. These packets are combined into carrier files for slingshot powered bouncing transfers and when the written file is accessed and used to continue its journey, they are separated into separate packets.

[0219] Fig.24 The reading process of the ejection node (SLN) during the ejection process is described. One of the key points of this exemplary embodiment is to show the ordering of file pull batches, the processing of carrier files of group / combined groupings or other files or data structures, and the processing of carrier files in parallel streams in an efficient manner to avoid duplication of received files.

[0220] This exemplary embodiment illustrates the importance of tick granularity in practical applications. It is based on the results from U.S. Provisional Patent Application No. 62 / 266,060 and International Patent Application No. PCT / IB16 / 01867. Fig.25 , and describes pulling a batch of files from a queue where new files will continue to appear.

[0221] exist Fig.24 In the example, there are delayed timed batches of files 24-5200, 24-5210, and 25-5220, and a mechanism to read and use only complete files. Those that are not yet complete (meaning they have not yet arrived completely) are passed on for inspection by subsequent read batches. Those that have already been used 24-5202, 24-5212, and 25-5222 are marked as read, and either moved to another folder, or deleted, or otherwise marked or processed so that the already processed files are not processed again by subsequent read queue batches.

[0222] Fig.24 Depicts a batch pull of files from PFS24-201 over a period of time starting at 24-5000 and ending at 24-5999. Time is depicted on two axes, time intervals on the top and batches on the left.

[0223] Interval A 24-5100 and Batch A Pull A 24-5200 occur simultaneously. Use 24-5202 occurs at the end of the interval. Interval B 24-5110 and Batch B Pull B 24-5210 occur simultaneously. Use 24-5212 occurs at the end of the interval. Interval C 24-5120 and Batch C Pull C 24-5220 occur simultaneously. Use 24-5222 occurs at the end of the interval.

[0224] Having an interval start right after an interval ends causes a problem because there may not be enough time to mark a file as read even though it has been read by a previous batch. This problem is described in U.S. Provisional Patent Application No. 62 / 266,060 and International Patent Application No. PCT / IB16 / 01867. Fig.24 This is a very dangerous flaw because, for example, executing a trade request twice could result in significant financial losses and / or other unintended consequences.

[0225] Fig.24 Similar in all respects to that disclosed in U.S. Provisional Patent Application No. 62 / 266,060 Fig.24 and International Patent Application No. PCT / IB16 / 01867 Fig.25 Similar, except that the figure includes a delay A 24-5102 between interval A 24-5100 and interval B 24-5110 (and between batch pull A 24-5200 and batch pull B 24-5210).

[0226] There is also a delay B 24-5112 between interval B 24-5110 and interval C 24-5120 (and between batch pull B 24-5210 and batch pull C 24-5220).

[0227] The key point is that this delay allows the current batch to evaluate and process all the files it has pulled, and mark those files as read if it has consumed the complete file.

[0228] This delay added to the mechanism is completely dynamic and can be extended if the batch requires more processing time, or shortened if the batch is complete. The interval time can also be adjusted dynamically based on several factors.

[0229] Fig.25 The timing sequence of reading files by the ejection node is described in more detail. If the files in the PFS incoming folder 25-606 are complete and when complete files are processed and incomplete files are ignored, the files are checked in the read queue process RQP (such as 25-RQP00 or 25-RQP10). The processed files are marked or otherwise processed in the post-process Q (such as 25-Q00 and 25-Q10). The synchronization phase is a delay or other form of time or other migration, such as 25-R00 and 25-R10 between batches.

[0230] The diagram shows a fixed time period for each batch, with a variable synchronization phase to help keep timing accurate. It also controls the maximum number of files that a batch can process if more files are to be processed than the expected processing and post-processing time.

[0231] Equation 25-1: Δt = P + Q + R.

[0232] t = the difference in time from the start of the tick to the end of the tick

[0233] P = the time taken by this tick batch item in the cycle.

[0234] Q = time used for post-batch processing.

[0235] R = Delay time to ensure there is no overlap between batch items and / or to set an offset between ticks.

[0236] Tick ​​granularity has been discussed in U.S. Provisional Patent Application No. 62 / 296,257. The tick granularity is based on several factors. The start time of a tick is based on the completion of the previous cycle of the tick call or according to a scheduled fixed time interval.

[0237] There are two types of cycles. Fixed time cycles set a finite number of items based on an estimate of the time it takes to process (P) each item and then handle post-processing (Q). This limit ensures that each cycle does not run longer than the allotted allowed time. Variable time cycles allow all items to be processed in that batch. The R delay time ensures that the next cycle pulls a new batch and all items in the last processed batch have been processed (P) and the post-processing (Q) has been completed. For example, for files in a queue - the list can be processed in the P stage, and then the files can be marked, moved, deleted, or touched by other means so that they are not pulled in the next batch. The time delay (R) ensures that there is enough delay between the last item processed and post-processed in the previous batch and the first item to be processed in the next batch.

[0238] The tick manager component maintains a history of runtime data for each cycle. These logged records can be used to analyze periods of under-utilization and peak usage. It can also detect delays due to unprocessed items being truncated due to a limited number of items per cycle. If this happens too often, then it is an indicator that more processing power is needed, the software is inefficient, or it may indicate other problems such as a slow database or the need for software refactoring, and / or other issues.

[0239] The granularity of the tick can be as fine as the application requires and can be set to run at fixed intervals, or at variable intervals based on period length. This tick granularity allows for consistent network beacon flashing, slingshot related functions, and other time-dependent functions. Further features and details on the tick granularity are presented in U.S. Provisional Patent Application No. 62 / 296,257.

[0240] This example shows two batch file pulls on a backbone exchange server (SRV_BBX) with read queue + process 25-RQP00 and read queue + process 25-RQP10. They both pull files from the same storage medium parallel file system PFS incoming file 25-606. The files pulled in 25-RQPOO via path 25-RQP606 are processed, and then in post-process P25-Q00, the files are marked via path 25-Q606.

[0241] This is a very important point because the next batch file pull read queue + process 25-RQP10 from the PFS incoming files 25-606 via path 25-RQP 616 should only include unmarked files or files that were not filled by the previous batch. Then in the post-P 25-Q10, the pulled and used files are marked via path 25-Q616 so that they are not inadvertently pulled by subsequent batch file pulls.

[0242] Fig.26 Batches with variable intervals (not fixed times) are shown. The times 26-T00 and 26-T10 and 26-T20 are based on the number of items processed in each batch, and the next batch will start shortly thereafter. Fig.26 Similar to Fig.25 However, it also illustrates the processing of a variable number of files and how this affects processing and post-processing times, and indicates the minimum delay before the next cycle should be executed.

[0243] Three cycles are indicated here: 26-T00, 26-T10 and 26-T20, and each cycle has a different number, namely 58 in 26-QNTY00, 22 in 26-QNTY10 and 8 in 26-QNTY20.

[0244] These different quantities affect the duration of each interval.

[0245] Fig. 27 based on Fig.24 And continues to describe batch processing of files, where the individual files shown are added. The use of files shows the use 5202 of files in A batch pull A 27-5200, the use 27-5212 of files in B batch pull B 27-5210, and the use 27-5222 of files in C batch pull C 27-5220.

[0246] Files File 27-00 and File 27-02 are sent in parallel using File (Batch A) 27-5206 and end at end of batch 27-5208. Files File 27-04 and File 27-06 and File 27-08 are sent in parallel using File (Batch B) 27-5216 and end at end of batch 27-5218. Files File 27-10 and File 27-12 and File 27-14 are sent in parallel using File (Batch C) 27-5226 and end at end of batch 27-5228.

[0247] The key point is that the pulling of files is done in parallel batches as they become available, so the use phase is also parallel. Files that are not ready in batch pull A 27-5200, such as file 27-04 and file 27-06, are not pulled and used until they are available. Files 27-04 and 27-06 are pulled along with file 27-08 in batch pull B 27-5210.

[0248] Fig.28 An algorithm is described to evaluate which is the more efficient transmission method, Delay 28-D10 for Internet channel measurements or Slingshot 28-D20.

[0249] Algorithmic decision logic that determines whether it is faster and / or more efficient to send packets via an Internet path or an Internet path enhanced by Slingshot with a Slingshot transparent segment between a pair (or more than one pair) of IP addresses. It does not consider reliability or loss or related congestion or other issues other than latency that may affect the performance of the Internet path.

[0250] Using bounces of Δt at each bridgehead of a bounce transparent relay segment is inefficient compared to the gain achieved via the use of bounces. A minimum distance is required to compensate for the necessary processing time of the bounce mechanism. The measured delay of the IP 28-D10 uses equation 13-1 to measure the total Internet path time Δt. The measured delay of the SL28-D20 uses equation 13-2 to measure the total Internet-Bounce Hybrid Path Time Δt.

[0251] IP to SL delay comparison 28-D30 uses equation 14-1 Δt{TV+TZ+TV}<Δt{TW ms}. There is a decision "IP to SL delay comparison 28-D30" which looks at the delay of the Internet path 28-CPT340 in the measurement of ms 28-TW ms to determine if it is higher or lower than the value of the sum of the time measurements of the slingshot path 28-CPT302 / 28-CPT306 time values ​​28-TV+28TZ+28-TV.

[0252] If it is true that the bounce-enhanced path (Δt{TV+TZ+TV}) has lower latency, SL is 28-DP50 lower than the Internet path, then “SL is the most efficient transport, using” 28-D50 is the optimal path

[0253] If false and the Internet (Δt{TW ms}) is faster than the bounce, "IP is lower" 28-DP40, then "IP is the most efficient transport, using" 28-D40 is the optimal path

[0254] Fig.29 Modules running on various devices are described to power Bounce integrated into the GVN. This exemplary embodiment shows a stack of four devices, namely an End Point Device (EPD) 100, an Access Point Server (SRV_AP) 300, a Central Control Server (SRV_CNTRL) 200, and a Backbone Switch Server (SRV_BBX) 500.

[0255] In SRV_BBX 500, it may have a bounce router device between it and the Internet path. In this topology, SRV_BBX is the infrastructure between the Internet and the backbone, using two reverse mechanisms. The slingshot router can act as a path enabler for bounce. For example, in an Internet Data Center (IDC), there can be a series of bounce routes as the front. This mechanism can be configured to bounce client devices. Slingshot can be managed by the service provider and the bounce router through the client.

[0256] Some key elements have been emphasized. There may be more elements that are not mentioned. Some of the elements mentioned are not directly affected by the bounce, rely on the bounce or otherwise integrated with the bounce, but have been mentioned to show the position in the stack where the project can be placed. The hierarchy and placement of the project can indicate the level of the element near the top as a high-level project, and the project as a lower-level project at the bottom. For example, network interface card (NIC) S108, S308, S208 and S508 are all at a very low system level. Operating system (O / S) S110, S310, S210 and S510 are above the NIC level, and there is a driver file connected to the NIC and operating the NIC inside the O / S. Some elements mentioned (and other elements that are not mentioned) may be at an appropriate level relative to other elements, or they may need to be lower or higher, depending on use, environment and other factors.

[0257] Other elements of GVN, Slingshot, Bounce or other related technologies include Fabric Manager, Log, AI, Security, FW, Secure Boot Manager (SBM), Back Channel Mechanism (BCM), Geographic Destination (Geo-D), Resource Manager, GVN Module, APP, Advanced Smart Routing (ASR), GVN Manager, Accounts, etc.

[0258] Slingshot Manager manages the hop listeners, file cache modules (receive), file cache managers (send), hop routers, file senders, and other items.

Claims

1. A method for determining an optimal path between endpoint devices, include: obtaining an Internet path time for transmitting the first packetized traffic from the first endpoint device to the second endpoint device via the Internet path; Obtaining a bounce enhanced path time for transmitting the first packetized traffic through a bounce enhanced path, wherein the bounce enhanced path time includes: a first time for a first switching server at the first endpoint device side to receive the first packetized traffic, convert the first packetized traffic into a carrier file, and write the carrier file into a first storage node, and a second time for a second switching server at the second endpoint device side to read the carrier file from the first storage node, convert the carrier file into a second packetized traffic, and transmit the second packetized traffic to the second endpoint device; Determining whether the Internet path time is less than the bounce enhancement path time; and In a case where the Internet path time is greater than the bounce enhancement path time, the bounce enhancement path is determined as the optimal path.

2. The method according to claim 1, further comprising: include: In a case where the Internet path time is less than the bounce enhancement path time, the Internet path is determined as the optimal path.

3. The method according to claim 1, further comprising: include: The third packetized traffic is transmitted through the bounce enhanced path.

4. The method according to claim 1 or 2, further comprising: include: The first packetized traffic is received by the first endpoint device from a third endpoint device.

5. The method according to claim 1 or 2, further comprising: include: The second packetized traffic is sent by the second endpoint device to a third endpoint device.

Images