Intelligent adaptive transport layer to enhance performance using multiple channels

Abstract

A set of connections is established, continuously evaluated and maintained between two endpoints of a computer network for use in transmitting information flows in a more efficient and controlled manner. New connections are established and existing connections are terminated in a continual search for connections with better and / or different performance characteristics. Each connection may utilize the same or a different path through the network and may have performance characteristics that change over time. Several paths can be used simultaneously for a given information flow to improve network metrics including: throughput, transaction time, data consistency, latency and packet loss. Flows of information can be broken into one or more sub-flows and sub-flows can be assigned to one or more active connections. Furthermore, dynamic decisions regarding how flows are broken up and how they are assigned to connections can be made in response to network conditions. Through the use of these connections, a reduced cost can be offered and application QoS / QoE can be guaranteed, allowing existing networks such as the public Internet to provide an enterprise class connection, which can be used to accelerate enterprise cloud adoption without modifying the present Internet infrastructure.

Description

RELATED APPLICATIONS[0001]This application is a continuation of U.S. patent application Ser. No. 15 / 626,130, filed on Jun. 18, 2017, now U.S. Pat. No. 10,868,752, issued on Dec. 15, 2020, which claims priority to U.S. Provisional Application No. 62 / 351,953, filed on Jun. 18, 2016, both of which are incorporated herein by reference.BACKGROUND[0002]This invention relates to the field of computer networking, and more specifically to controlling network metrics such as latency, flow completion time (FCT) and throughput between endpoints of a packet-switched network. This includes networks such as the public Internet, private networks, and 3G / 4G / 5G mobile networks.[0003]The Internet provides excellent connectivity while ensuring properties such as resilience, decentralization, and best effort packet delivery. However, these characteristics result in low utilization of the Internet core to handle peaks of traffic. In addition, the Internet is not generally deterministic, a fact that inhib...

AI Technical Summary

Benefits of technology

The patent text discusses the problem of managing network metrics, such as latency and throughput, in a way that ensures the reliability and quality of service of critical applications while also being cost-effective for non-critical applications. The current options for private networks are expensive and not accessible to all enterprises. The invention aims to provide an improved method of controlling network metrics while guaranteeing quality of service and scalability. The text also highlights the challenges of managing network metrics in the context of cloud computing, where data is generated and consumed in different locations and the communication between enterprises and datacenters is critical. The technical effects of the invention include offering a network capable of managing critical applications, ensuring certain boundaries for non-critical applications, and providing low-cost and excellent scalability while maintaining reliability.

Problems solved by technology

In addition, they operate on a per-link basis since the visibility of each router is limited to its neighbors.

This architectural decision greatly contributes to the Internet scalability but penalizes its performance for other metrics such as throughput and latency.

In addition, new problems arise such as lack of control over the data.

In an alternative embodiment, the tunnels are not secure to reduce the computation time required to generate encrypted packets and increase the performance.

The increased reliability comes at a cost.

This technique has a large overhead since the same traffic sent multiple times interferes with each other and congests the routers.

This technique reduces the congestion but can potentially aggravate the out of order problem.

Both techniques share a common overhead to create the multiple paths.

For instance, the overhead cost of MPTCP may counteract its advantages in the transfer of small files.

Unlike TCP, UDP is not a reliable transport protocol, i.e., it leaves to the application the issues of dropped, out of order and duplicated packets.

Packet Loss: In a packet-switched system, packet loss refers to the number of packets that fail to arrive at their intended destination.

The main factors that cause packet loss are link congestion, device performance (router, switch, etc.) such as buffer overloads, software issues on network devices, and faulty hardware.

This can be limited by different things including latency, packet loss, and what protocol is being used.

Latency in packet switched networks can be affected by many different factors, especially in the operating environment of long distance networks, such as processing delay, buffer bloat and queueing delays.

Jitter results from network congestion, timing drift and route changes.

As reported in Internet Engineering Task Force (IETF) Request for Comments (RFC) 3393, incorporated herein by reference, the term jitter in packet-switched networks to identify the variation in packet delay is not completely correct.

Although this is a simple network topology, enterprises may require more complex topologies to interconnect their offices.

Also, the number of flows is limited by the capacity of the CSPs, which is the bandwidth of each connection.

This results in each direction of a tunnel potentially having different characteristics (e.g., in terms of latency, or bandwidth).

However, some applications and / or protocols may suffer from this asymmetry in the characteristics of both flow directions.

This restriction stems from the number of available transport ports publicly open by the endpoint, as there are security concerns regarding opened ports.

Running a VM in each device may not be desirable by some companies because they prefer to use a firewall to secure their LANs.

While running VMs on all devices results in an end-to-end QoE, running them in egress / ingress points of LANs may result in performance uncertainties because QoE is guaranteed only between those two points.

Furthermore, CSPs inside the company likely share most of their links since the internal routing options are typically limited.

This fact can impact performance due to cross-interference and congestion.

Despite this change, it is not guaranteed that the “new” tunnel has different properties than the discarded or active tunnels.

The traceroute method is not entirely accurate.

First, routers do not have the obligation to answer, and in case they do they may not send the correct information about themselves.

While this can be done on purpose, this is not generally the case.

Also, the traceroute technique makes an assumption that is not always satisfied in packet-switching networks.

In this case, incremental TTL packets do not provide the router identification within the same path.

This fact could lead to potential incorrect link identifications.

However, this technique has drawbacks that affect all tracerouting techniques: (i) NAT addresses rewriting in private networks, and (ii) ISP non-responding when a TTL expires.

A specific drawback peculiar to the modified traceroute technique is the possibility of not receiving any answer from the last hop as the 5-tuple is an actual valid tuple that is forwarded to a listening service on that port (e.g., the OpenVPN service).

For instance, a waiting CSP may have high performance sometimes and poor performance during other periods.

This topology could result in increased complexity of the policies.

A shorter interval gets a finer granularity at a cost of a huge overhead both in processing and transmission efforts.

Thus, there is trade-off between accuracy and resource usage to control the system's granularity.

The control in this case is off-loaded from the data layer due to scalability issues.

Another more restrictive situation would be to use only a specific active CSP.

Here, the data layer cannot optimize hardly anything because the policy constrains the available behaviors.

The amount of data to consider is large and it greatly varies over time, a fact that poses significant challenges to the architecture.

Images