A method and apparatus for providing RDMA dynamic connection transport services

Abstract

This invention provides a method and apparatus for providing RDMA dynamic connection transmission services. The method includes: receiving a dynamic connection request message from RDMA; extracting the source port network address and the send queue identifier of the dynamic connection requester from the packet header and performing a hash operation; querying the corresponding hash table entry based on the hash value to match the dynamic connection context resource; if the match is successful, generating a context resource identifier corresponding to the dynamic connection context resource based on the hash value and the corresponding hash table entry identifier, writing the context information into the corresponding context on-chip buffer, and transmitting the dynamic connection request message to the RDMA receiving engine; if the match fails, silently discarding the abnormal message, and transmitting a data request message after successful re-establishment of the connection. This invention improves resource utilization and effective bandwidth, and enhances the robustness of the system.

Description

Technical Field

[0001] This invention belongs to the field of network transmission, and specifically relates to a method and apparatus for providing RDMA dynamic connection transmission services. Background Technology

[0002] Similar to DMA technology within computer systems, RDMA (Remote Direct Memory Access) technology supports zero-copy network technology and kernel bypass technology in high-performance computing and large data centers, enabling high-performance direct remote data access. The InfiniBand protocol (IB protocol for short) is a network protocol that supports RDMA technology. The IB protocol reinterprets an entire network layer model, including the transport layer, network layer, and link layer, making common network interface cards (NICs), switches, and other devices support RDMA technology. In RDMA technology, there are three commonly used reliable connection service types: Reliable Connection (RC), Extended Reliable Connection (XRC), and Dynamic Connection (DC). DC service, due to its ability to dynamically establish and release connections, is more flexible in its connection methods and consumes fewer resources. DC establishes connections using the connection management methods defined in the IB protocol, i.e., establishing a communication channel and exchanging context information. Its connection establishment process does not rely on connection management methods but instead notifies the destination end of connection establishment through hardware packet assembly. Similarly, to release a connection, a disconnection request message is sent. This special packet is called a DC connection packet.

[0003] The specific process of DC chain establishment and disconnection is as follows: Figure 1 As shown, it specifically includes:

[0004] ① The software creates the DC service initiator requester (DCI) and the DC service destination receiver (DCT) according to the configuration. Each DCI corresponds to a sending queue of an entity. This sending queue is the same as the requester of the RC service type and contains some basic information. The DCT is the destination object for receiving and processing DC operations, and each DCT corresponds to a shared receiving queue.

[0005] ② The software constructs a sufficient number of receive work requests (WR) in the shared receive queue at the receiving end. Each WR request consists of three important information segments: the ctrl segment and the data segment contain basic information and the address length of the data to be moved, etc.; the target segment is the basis used by the sender to parse the destination end, and the data segment specifies a buffer of the same size to ensure that requests initiated by multiple requesting ends and multiple DCIs can be processed.

[0006] ③ When the upper-layer application needs to move data, it sends a WR request to the sending queue.

[0007] ④ The sending controller compares it with the information in the current DCI context. If the information matches, it continues to process the WR. If they do not match, it disconnects the current DCI connection and re-establishes a connection with the target pointed to in the WR.

[0008] ⑤ The receiver controller determines whether the received request is a data request or a connection establishment request. If a connection establishment request is received, it allocates an idle DCR (Data Receiver Context) from the resource pool associated with the DCT to establish a connection with the DCI and store information. After allocation, it returns a response packet. If a data request is received, the receiver controller finds the corresponding DCR based on the information in the packet and processes the data.

[0009] DC transport provides a datagram-like model that allows a process to connect to multiple processes on multiple remote nodes, binding contexts together through network data request messages to create a temporary equivalent RC transport service. This dynamic connection management approach means that any created DCI may repeatedly establish and terminate connections during its lifecycle. During connection establishment, unused DCRs may be released, and newly enabled DCRs may be put into active status.

[0010] Currently, the typical RDMA DC transmission service technology was proposed by Mellanox, and its data transmission ladder diagram is as follows: Figure 2 As shown, DC connection establishment request messages and data request messages are sent back-to-back. The receiver uses the source DCI identifier and source port network address (LID) carried in the DC connection establishment request message to query the allocated DCR. Upon successful allocation, an ACK response is returned; if no resources are available to process the connection request, a CNAK connection rejection response is returned. If a DC disconnection request message is received, the responder does not need to return any response; the sender assumes the disconnection was successful.

[0011] With the improvement of chip manufacturing technology, traditional DC implementation schemes have some shortcomings when dealing with the demanding requirements of lossless network transmission. Specifically, these shortcomings are as follows:

[0012] 1) High complexity in handling connection context

[0013] On the requesting end, each process typically maintains one or more DC contexts. The hardware network interface card (NIC) transfers these context data structures from host memory to its context cache as needed. In scenarios where multiple nodes use the DC transport service type to initiate a connection request, the receiver of the target node is determined jointly by the source requester's DCI identifier and the requester's LID. However, the DCI identifier is 24 bits of data, and the LID is 16 bits. Mapping these two together results in low query efficiency and relatively low storage space utilization.

[0014] 2) DC transmission is highly independent, but difficult to manage.

[0015] Because the DC transport service allows a datagram-like connection model on the connection entity, the DCI and DCR are implemented as independent connection contexts, not bound in pairs. When a DCI is bound to a DCR on a given remote node, it does not mean there must be a complementary connection in the opposite direction simultaneously. Therefore, the DC transport service differs fundamentally from services like RC in its connection management mechanism, necessitating separate sender and receiver controllers for DC transport within the system framework, impacting system design complexity and processing stability.

[0016] 3) Large delay in handling connection packet anomalies

[0017] Compared to RC transmission, DC transmission adds connection semantics such as connection establishment and termination. Correspondingly, when handling abnormal situations such as packet loss, sequence disorder, state errors, and resource insufficiency, the connection packet status needs to be considered. Currently, when processing connection establishment request messages, if the responder has insufficient resources (e.g., all DCRs in the resource pool are in use), it will reject the connection establishment request and reply with a CNAK. The responder does not retain any information on the rejected connection request, and subsequent data request messages will not match any active DCRs; other requests will be silently dropped. Upon receiving a CNAK, the DCI will wait for a configurable period of time before retrying. Since connection establishment request messages and data request messages are sent back-to-back, the sender assumes successful connection establishment before receiving an ACK response to the connection establishment request message. If a CNAK occurs at this point, the data request message will be silently dropped by the receiver. After retransmission, if no resources are available, the data request message will be dropped again, affecting effective bandwidth.

[0018] Furthermore, a broken link request message can only be sent after all data request messages have returned ACK responses. If a broken link request message is lost and the DCR is not successfully released, when the same DCI needs to make another request to the same DCT, the DCR may have already been allocated but the information may not be up-to-date. This will eventually cause the receiver to receive a valid request but identify it as an invalid request, leading to an error state. The host software will then need to reset the send / receive queue, affecting normal functionality. Summary of the Invention

[0019] The purpose of this invention is to provide a method and apparatus for providing RDMA dynamic connection transmission services, which aims to realize RDMA DC transmission services and overcome the limitations of DCR context management and the defects of connection packet exception handling.

[0020] According to a first aspect of the present invention, a method for providing RDMA dynamic connection transport services is provided, comprising:

[0021] Upon receiving a dynamic connection request message, extract the source port network address and the sending queue identifier of the dynamic connection requester from the packet header and perform a hash operation. Based on the hash value, query the corresponding hash table entry in the hash table to match the dynamic connection context resource.

[0022] If the dynamic connection context resource is successfully matched, a context resource identifier corresponding to the dynamic connection context resource is generated based on the hash value and the corresponding hash table entry identifier, the context information is written into the context on-chip buffer, and the dynamic connection request message is transmitted to the RDMA receiving engine.

[0023] If the dynamic connection context resource matching fails, it is determined that the dynamic connection request message is abnormal. The abnormal message is silently discarded, and the data request message is transmitted after the connection is successfully re-established.

[0024] Preferably, before receiving the dynamic connection request message, the method further includes:

[0025] Determine whether the received message is a dynamic connection request message. If it is not a dynamic connection request message, pass it directly to the RDMA receiving engine for processing.

[0026] Preferably, the step of querying the corresponding hash table entry in the hash table based on the hash value further includes:

[0027] The source port network address and the sending queue identifier of the dynamic connection requester are hashed, and the hash value is used as the address. The hash table entry corresponding to the address is then searched.

[0028] If the dynamic connection request message is a link establishment request message, then when an available hash table entry exists, the source port network address and the sending queue identifier of the dynamic connection requester are filled into the hash table entry, and it is determined that the dynamic connection context resource is successfully matched; otherwise, the link establishment request is rejected.

[0029] Preferably, the context resource identifier consists of a dynamic connection transport service fixed offset value, the hash value, and a hash table entry identifier.

[0030] Preferably, transmitting the data request message after successful re-establishment of the link further includes:

[0031] When an anomaly occurs in the message, a connection rejection response message is returned to the dynamic connection transport service requester, notifying the dynamic connection transport service requester to retransmit the connection establishment request message.

[0032] After the connection is successfully established, an acknowledgment response is returned to the dynamic connection transmission service requester, who then initiates a data request message.

[0033] According to a second aspect of the present invention, an RDMA dynamic connection transport service providing apparatus is provided, comprising:

[0034] The resource matching unit is used to receive dynamic connection request messages, extract the source port network address and the sending queue identifier of the dynamic connection requester from the packet header, perform a hash operation, and query the corresponding hash table entry in the hash table according to the hash value to match the dynamic connection context resources.

[0035] The cache writing unit is used to generate a context resource identifier corresponding to the dynamic connection context resource according to the hash value and the corresponding hash table entry identifier when the dynamic connection context resource is successfully matched, write the context information into the context on-chip cache, and transmit the dynamic connection request message to the RDMA receiving engine.

[0036] An abnormal message processing unit is used to determine that the dynamic connection request message is abnormal when the dynamic connection context resource matching fails, silently discard the abnormal message, and transmit the data request message after the connection is successfully re-established.

[0037] The RDMA dynamic connection transmission service provision method and apparatus of the present invention can solve the defects of Mellanox in implementing RDMA dynamic connection transmission services, such as dealing with mixed DC and RC transmission service scenarios and problems such as insufficient receiver resources or packet loss anomalies. Compared with the prior art, it has the following advantages:

[0038] 1) This invention performs hash operations on LID and DCIN, accelerating the query process for available DCR resources. The mechanism of returning DCRn via ACK packets further improves query efficiency, reduces redundant hash calculations, and allows for more direct location of the associated DCR context. Furthermore, management through a shared DCR resource pool avoids uneven resource distribution caused by varying resource occupancy among different DCTs, reduces the frequency of CNAK occurrences, and improves resource utilization, enabling more effective use of limited DCR resources.

[0039] 2) The solution of this invention writes the DCR connection context generated after processing the DC connection request message into the Context Cache context on-chip cache for management, along with contexts of other transport service types. The generated DCRn is used as the access identifier for the corresponding context. The fixed offset value of the DC transport service in the DCRn identifier also avoids duplication of the generated identifier with other transport service types. This solution simplifies the design, allowing other modules to process the data directly according to the normal processing flow without needing to be aware of the specific service type, thus facilitating management and maintenance while fulfilling functional requirements.

[0040] 3) This invention optimizes the handling of CNAK retransmission and DC link break request packet loss. When a retransmission is initiated after waiting for a CNAK, the DC data request packet is sent only after receiving the ACK response packet for the DC link establishment request packet, instead of transmitting the DC data request packet back-to-back with the DC link establishment request packet. This effectively avoids the situation where the DC data request packet is silently discarded when there are no available resources and a CNAK is replied to again, thus improving the effective bandwidth. In addition, this invention also implements configurable ACK response for DC link break request packets. When there is an application requirement for repeated link establishment and breakage for a certain DCT in the WR sent to the DCI, this configuration option can be enabled to avoid abnormal errors caused by the DC link break request packet failing to reach the destination for some reason, thus enhancing the robustness of the system.

[0041] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures and processes pointed out in the description, claims, and drawings. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 It is a flowchart of RDMA DC connection management based on existing technology.

[0044] Figure 2 It is a ladder diagram for data transmission based on existing DC schemes.

[0045] Figure 3 This is a flowchart of the RDMA dynamic connection transmission service provision method according to the present invention.

[0046] Figure 4 This is an architecture diagram of the implementation scheme of the RDMA DC transmission service type according to the present invention.

[0047] Figure 5 This is a ladder diagram of the CNAK retransmission scheme and the link break and packet loss scheme according to the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] To address the limitations of DCR context management and the deficiencies in connection packet anomaly handling in Mellanox's implementation, this invention proposes a method and apparatus for providing RDMA dynamic connection transmission services. Based on the traditional RDMA network implementation, a DCT management module is added to the receiver at the destination end to parse received request packets. A hash group management system is used for the DCR resource pool within a node. Idle DCRs in the pool are not bound to specific DCTs but are treated as shared resources. After identifying a DC connection packet, a corresponding hash group is calculated based on the DCI identifier and source port network address (LID) carried in the packet header. Idle DCR resources are then identified and allocated within this hash group. When allocating DCRs, the destination DCT context information is read, and the created DCR context is combined with the packet header information and written to the context on-chip cache, ensuring consistency with RC in context management. If resource shortages occur, the DCT management module generates an error code and forwards it to the downstream module, returning a CNAK packet to the requesting end. Furthermore, by extending the handling scheme for DC connection packets in abnormal scenarios, rapid anomaly recovery is achieved. This solution aims to improve resource utilization and enhance system processing capabilities.

[0050] See Figure 3 The flowchart illustrates that the RDMA dynamic connection transmission service provision method provided by this invention includes:

[0051] Step 101: Receive the RDMA dynamic connection request message, extract the source port network address and the dynamic connection requester's send queue identifier from the packet header and perform a hash operation, query the corresponding hash table entry based on the hash value, and match the dynamic connection context resource.

[0052] The schematic diagram of the implementation architecture of this invention is as follows: Figure 4As shown. The overall architecture consists of a DCT management module and standard RDMA components, where the DCT management module is DCT_MAN, and the standard RDMA components are the receive engine RX_ENG and the on-chip context cache.

[0053] The DCT_MAN module receives request messages and parses them based on the operation type field in the packet header. If it is not a DC request message, it is directly passed to the RX_ENG receiving engine for normal processing; otherwise, it is processed by DCT_MAN.

[0054] The DCT_MAN module includes a hash table for managing the DCR resource pool within a node using hash groups. When a DC retrieval request message is received, the source port network address (LID) and the DC requester's sending queue identifier (DCIN) in the packet header are hashed. The resulting hash value is used as the address to query the available DCR resources (hash table sub-entries) in the hash table. Each hash value corresponds to multiple (m sub-entries shown in the diagram) to prevent hash collisions.

[0055] Step 102: If the dynamic connection context resource is successfully matched, then generate the context resource identifier corresponding to the dynamic connection context resource according to the hash value and the corresponding hash table entry identifier, and write the context information into the context on-chip cache.

[0056] The DCT_MAN module performs different operations depending on whether the DC request message is a link establishment request message, a link termination request message, or a data request message.

[0057] If the DC request message is a link establishment request message, find an available sub-entry (with field vld set to 0) in the HASH Table, fill in the LID and DCIN into the sub-entry, and set field vld to 1. If no available sub-entry is found, reject the link establishment request. If the link establishment is successful, generate a DCR identifier DCRn based on the HASH value and the identifier of the sub-entry.

[0058] DCRn = {DC transmission service fixed offset value, HASH value, sub-item identifier}.

[0059] The DC transmission service fixed offset value is used to avoid the generated identifier DCRn being duplicated with other transmission service types.

[0060] The generated DCRn is fed back in the ACK message, and the DCT context is read according to the DCT identifier in the packet header. Combined with the valid field information in the packet header, it is written into the storage space corresponding to DCRn in the Context Cache context on-chip cache, so that it is consistent with RC in terms of context management.

[0061] In an optional embodiment, if the DC request message is a disconnection request message, then the HASH Table is searched for a sub-item that matches LID and DCIN and has vld of 1, and its vld is set to 0, indicating that the sub-item is in an available state.

[0062] Step 103: If the dynamic connection context resource matching fails, it is determined that the dynamic connection request message is abnormal. The abnormal message is silently discarded, and the data request message is transmitted after the connection is successfully re-established.

[0063] If the DC request message is a data request message, first determine if it carries a DCR identifier (DCRn). If it doesn't carry the DCRn returned by the ACK, then query the corresponding sub-item based on the hash values ​​of LID and DCIN to see if it's valid and matches. If found, generate the DCRn in the same way as the link establishment message; otherwise, it's considered a DC abnormal request message. On the other hand, if it carries the DCRn returned by the ACK, then index the corresponding sub-item based on the DCRn. If it's invalid or the information doesn't match, it's considered a DC abnormal request message.

[0064] If the message is a normal DC request or a message indicating a refused connection, it is passed to the RX_ENG receiving engine for further processing. If the message is an abnormal DC request, it is discarded directly. This invention expands the handling scheme for DC connection packets in abnormal scenarios, eliminating the impact of silently discarding data request messages after CNAK retransmission that reduces effective bandwidth, and eliminating the potential for sending and receiving to enter error states due to packet loss of disconnection request messages.

[0065] like Figure 5 As shown in (a), when sending DC link establishment request messages and data request messages REQ normally, they can be transmitted back-to-back to improve the transmission rate. However, if the DC link establishment request message cannot find an available DCR resource, it returns CNAK to the requester. After the requester retransmits the DC link establishment request message, it will not immediately send a data request message REQ. Instead, it will wait for the ACK packet returned after the link establishment is successful before initiating the data request message REQ. This avoids the problem of reduced effective bandwidth caused by the silent discarding of data request messages.

[0066] like Figure 5As shown in (b), in application scenarios with frequent connection switching and a switchback to a previously requested DCT, a waiting-for-disconnection-ACK state (marked by the dashed box in the diagram) is added. Subsequent packets can only be sent after the ACK response to the disconnection request message is returned, effectively adding an isolation layer between the disconnection request message and the connection establishment request message for the next destination. Since the disconnection request message should originally only be sent after the ACK responses to all data request messages are returned, waiting for the ACK response of the disconnection request message before sending the connection establishment request message for the next destination avoids significant transmission delays and effectively prevents abnormal errors caused by the DC disconnection request message failing to reach the destination for some reason, thus improving the stability of the system design.

[0067] As can be seen, the RDMA dynamic connection transmission service provision method proposed in this invention can solve the shortcomings of Mellanox in implementing RDMA dynamic connection transmission services, such as dealing with mixed DC and RC transmission service scenarios and problems such as insufficient receiver resources or packet loss anomalies. Compared with the prior art, it has the following advantages:

[0068] 1) This invention performs hash operations on LID and DCIN, accelerating the query process for available DCR resources. The mechanism of returning DCRn via ACK packets further improves query efficiency, reduces redundant hash calculations, and allows for more direct location of the associated DCR context. Furthermore, management through a shared DCR resource pool avoids uneven resource distribution caused by varying resource occupancy among different DCTs, reduces the frequency of CNAK occurrences, and improves resource utilization, enabling more effective use of limited DCR resources.

[0069] 2) The solution of this invention writes the DCR connection context generated after processing the DC connection request message into the Context Cache context on-chip cache for management, along with contexts of other transport service types. The generated DCRn is used as the access identifier for the corresponding context. The fixed offset value of the DC transport service in the DCRn identifier also avoids duplication of the generated identifier with other transport service types. This solution simplifies the design, allowing other modules to process the data directly according to the normal processing flow without needing to be aware of the specific service type, thus facilitating management and maintenance while fulfilling functional requirements.

[0070] 3) This invention optimizes the handling of CNAK retransmission and DC link break request packet loss. When a retransmission is initiated after waiting for a CNAK, the DC data request packet is sent only after receiving the ACK response packet for the DC link establishment request packet, instead of transmitting the DC data request packet back-to-back with the DC link establishment request packet. This effectively avoids the situation where the DC data request packet is silently discarded when there are no available resources and a CNAK is replied to again, thus improving the effective bandwidth. In addition, this invention also implements configurable ACK response for DC link break request packets. When there is an application requirement for repeated link establishment and breakage for a certain DCT in the WR sent to the DCI, this configuration option can be enabled to avoid abnormal errors caused by the DC link break request packet failing to reach the destination for some reason, thus enhancing the robustness of the system.

[0071] Accordingly, in a second aspect, the present invention provides an RDMA dynamic connection transport service providing apparatus, comprising:

[0072] The resource matching unit is used to receive dynamic connection request messages, extract the source port network address and the sending queue identifier of the dynamic connection requester from the packet header, perform a hash operation, and query the corresponding hash table entry in the hash table according to the hash value to match the dynamic connection context resources.

[0073] The cache writing unit is used to generate a context resource identifier corresponding to the dynamic connection context resource according to the hash value and the corresponding hash table entry identifier when the dynamic connection context resource is successfully matched, write the context information into the context on-chip cache, and transmit the dynamic connection request message to the RDMA receiving engine.

[0074] An abnormal message processing unit is used to determine that the dynamic connection request message is abnormal when the dynamic connection context resource matching fails, silently discard the abnormal message, and transmit the data request message after the connection is successfully re-established.

[0075] The above-described apparatus can be implemented using the RDMA dynamic connection transmission service provision method provided in the first aspect of the embodiments. For specific implementation details, please refer to the description in the first aspect of the embodiments, which will not be repeated here.

[0076] It is understood that the hardware structures, parameters, and algorithm types described in the above embodiments are merely examples. Those skilled in the art can also make readily conceived combinations and adjustments to the structural features of the above embodiments according to usage needs, and the concept of the present invention should not be limited to the specific details of the above examples.

[0077] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for providing RDMA dynamic connection transmission services, characterized in that, include: Upon receiving a dynamic connection request message, extract the source port network address and the sending queue identifier of the dynamic connection requester from the packet header and perform a hash operation. Based on the hash value, query the corresponding hash table entry in the hash table to match the dynamic connection context resource. If the dynamic connection context resource is successfully matched, a context resource identifier corresponding to the dynamic connection context resource is generated according to the hash value and the corresponding hash table entry identifier, the context information is written into the context on-chip cache, and the dynamic connection request message is transmitted to the RDMA receiving engine. If the dynamic connection context resource matching fails, it is determined that the dynamic connection request message is abnormal. The abnormal message is silently discarded, and the data request message is transmitted after the connection is successfully re-established.

2. The RDMA dynamic connection transmission service provision method according to claim 1, characterized in that, Before receiving the dynamic connection request message, the method further includes: Determine whether the received message is a dynamic connection request message. If it is not a dynamic connection request message, pass it directly to the RDMA receiving engine for processing.

3. The RDMA dynamic connection transmission service provision method according to claim 1, characterized in that, The step of querying the corresponding hash table entry in the hash table based on the hash value further includes: The source port network address and the sending queue identifier of the dynamic connection requester are hashed, and the hash value is used as the address. The hash table entry corresponding to the address is then searched. If the dynamic connection request message is a link establishment request message, then when an available hash table entry exists, the source port network address and the sending queue identifier of the dynamic connection requester are filled into the hash table entry, and it is determined that the dynamic connection context resource is successfully matched; otherwise, the link establishment request is rejected.

4. The RDMA dynamic connection transmission service provision method according to claim 1, characterized in that, The context resource identifier consists of a fixed offset value for the dynamic connection transport service, the hash value, and a hash table entry identifier.

5. The RDMA dynamic connection transmission service provision method according to claim 1, characterized in that, The transmission of the data request message after successful re-establishment of the connection further includes: When an anomaly occurs in the message, a connection rejection response message is returned to the dynamic connection transport service requester, notifying the dynamic connection transport service requester to retransmit the connection establishment request message. After the connection is successfully established, an acknowledgment response is returned to the dynamic connection transmission service requester, who then initiates a data request message.

6. An RDMA dynamic connection transmission service providing device, characterized in that, include: The resource matching unit is used to receive dynamic connection request messages, extract the source port network address and the sending queue identifier of the dynamic connection requester from the packet header, perform a hash operation, and query the corresponding hash table entry in the hash table according to the hash value to match the dynamic connection context resources. The cache writing unit is used to generate a context resource identifier corresponding to the dynamic connection context resource according to the hash value and the corresponding hash table entry identifier when the dynamic connection context resource is successfully matched, write the context information into the context on-chip cache, and transmit the dynamic connection request message to the RDMA receiving engine. An abnormal message processing unit is used to determine that the dynamic connection request message is abnormal when the dynamic connection context resource matching fails, silently discard the abnormal message, and transmit the data request message after the connection is successfully re-established.

7. The RDMA dynamic connection transmission service providing apparatus according to claim 6, characterized in that, The resource matching unit is further configured to: Before receiving the dynamic connection request message, it is determined whether the received message is a dynamic connection request message. If it is not a dynamic connection request message, it is directly passed to the RDMA receiving engine for processing.

8. The RDMA dynamic connection transmission service providing apparatus according to claim 6, characterized in that, The resource matching unit is further configured to: The source port network address and the sending queue identifier of the dynamic connection requester are hashed, and the hash value is used as the address. The hash table entry corresponding to the address is then searched. If the dynamic connection request message is a link establishment request message, then when an available hash table entry exists, the source port network address and the sending queue identifier of the dynamic connection requester are filled into the hash table entry, and it is determined that the dynamic connection context resource is successfully matched; otherwise, the link establishment request is rejected.

9. The RDMA dynamic connection transmission service providing apparatus according to claim 6, characterized in that, The context resource identifier consists of a fixed offset value for the dynamic connection transport service, the hash value, and a hash table entry identifier.

10. The RDMA dynamic connection transmission service providing apparatus according to claim 6, characterized in that, The exception message processing unit is further used for: When an anomaly occurs in the message, a connection rejection response message is returned to the dynamic connection transport service requester, notifying the dynamic connection transport service requester to retransmit the connection establishment request message. After the connection is successfully established, an acknowledgment response is returned to the dynamic connection transmission service requester, who then initiates a data request message.

Images