Methods, systems and computer readable media for improving remote direct memory access performance
The subject matter described herein includes methods, systems, and computer readable media for improving remote direct memory access (RDMA) performance. A method for improving RDMA performance occurs at an RDMA node utilizing a user space and a kernel space for executing software. The method includes posting, by an application executing in the user space, an RDMA work request including a data element indicating a plurality of RDMA requests associated with the RDMA work request to be generated by software executing in the kernel space; and generating and sending, by the software executing in the kernel space, the plurality of RDMA requests to or via a system under test (SUT).
1. A method for improving remote direct memory access (RDMA) performance, the method comprising:
at a remote direct memory access (RDMA) client utilizing a user space and a kernel space for executing software:
posting, by an application executing in the user space, a single RDMA work request including a data element indicating a plurality of RDMA requests associated with the single RDMA work request to be generated by software executing in the kernel space, wherein the data element includes a work queue entry loop count parameter comprising a positive integer indicating the number of RDMA requests in the plurality of RDMA requests;
in response to the single RDMA work request and, using the work queue entry loop count parameter of the single RDMA work request, generating and sending, by the software executing in the kernel space, the plurality of RDMA requests to an RDMA server via a system under test (SUT), wherein the plurality of RDMA requests are identical, the SUT comprises at least one network switch located between the RDMA client and the RDMA server, and sending the RDMA requests to the RDMA server via the SUT includes stress testing the at least one network switch by flooding the at least one network switch with data packets carrying the RDMA requests; and
after determining, by the software executing in kernel space, that operations requested by the plurality of RDMA requests have been completed, posting, by the software executing in kernel space, a notification indicating that processing of the single RDMA work request is completed.
2. The method of claim 1 wherein determining that operations requested by the plurality of RDMA requests have been completed includes receiving a corresponding response for each of the plurality of RDMA requests.
3. The method of claim 1 wherein each of the plurality of RDMA requests is generated and sent without multiple context-switches between the kernel space and the user space.
4. The method of claim 1 wherein the RDMA work request is posted using an InfiniBand (IB) verbs related application programming interface (API).
5. The method of claim 1 wherein the plurality of RDMA requests includes a read operation request or a write operation request.
6. A system for improving remote direct memory access (RDMA) performance, the system comprising:
at least one processor; and
a remote direct memory access (RDMA) client implemented using the at least one processor, the RDMA node utilizing a user space and a kernel space for executing software, wherein the RDMA node is configured for:
posting, by an application executing in the user space, a single RDMA work request including a data element indicating a plurality of RDMA requests associated with the single RDMA work request to be generated by software executing in the kernel space, wherein the data element includes a work queue entry loop count parameter comprising a positive integer indicating the number of RDMA requests in the plurality of RDMA requests;
in response to the single RDMA work request and, using the work queue entry loop count parameter of the single RDMA work request, generating and sending, by the software executing in the kernel space, the plurality of RDMA requests to an RDMA server via a system under test (SUT), wherein the plurality of RDMA requests are identical, the SUT comprises at least one network switch located between the RDMA client and the RDMA server, and sending the RDMA requests to the RDMA server via the SUT includes stress testing the at least one network switch by flooding the at least one network switch with data packets carrying the RDMA requests; and
after determining, by the software executing in kernel space, that operations requested by the plurality of RDMA requests have been completed, posting, by the software executing in kernel space, a notification indicating that processing of the single RDMA work request is completed.
7. The system of claim 6 wherein determining that operations requested by the plurality of RDMA requests have been completed includes receiving a corresponding response for each of the plurality of RDMA requests.
8. The system of claim 6 wherein each of the plurality of RDMA requests is generated and sent without multiple context-switches between the kernel space and the user space.
9. The system of claim 6 wherein the RDMA work request is posted using an InfiniBand (IB) verbs related application programming interface (API).
10. The system of claim 6 wherein the plurality of RDMA requests includes a read operation request or a write operation request.
11. A non-transitory computer readable medium having stored thereon executable instructions that when executed by at least one processor of at least one computer cause the at least one computer to perform steps comprising:
at a remote direct memory access (RDMA) client utilizing a user space and a kernel space for executing software:
posting, by an application executing in the user space, a single RDMA work request including a data element indicating a plurality of RDMA requests associated with the single RDMA work request to be generated by software executing in the kernel space, wherein the data element includes a work queue entry loop count parameter comprising a positive integer indicating the number of RDMA requests in the plurality of RDMA requests;
in response to the single RDMA work request and, using the work queue entry loop count parameter of the single RDMA work request, generating and sending, by the software executing in the kernel space, the plurality of RDMA requests to an RDMA server via a system under test (SUT), wherein the plurality of RDMA requests are identical, the SUT comprises at least one network switch located between the RDMA client and the RDMA server, and sending the RDMA requests to the RDMA server via the SUT includes stress testing the at least one network switch by flooding the at least one network switch with data packets carrying the RDMA requests; and
after determining, by the software executing in kernel space, that operations requested by the plurality of RDMA requests have been completed, posting, by the software executing in kernel space, a notification indicating that processing of the single RDMA work request is completed.
12. The non-transitory computer readable medium of claim 11 wherein each of the plurality of RDMA requests is generated and sent without multiple context-switches between the kernel space and the user space.