mbox series

[v1,0/3] Optimize NFSD Send completion processing

Message ID 162731055652.13580.8774661104190191089.stgit@klimt.1015granger.net (mailing list archive)
Headers show
Series Optimize NFSD Send completion processing | expand

Message

Chuck Lever III July 26, 2021, 2:46 p.m. UTC
The following series improves the efficiency of NFSD's Send
completion processing by removing costly operations from the svcrdma
Send completion handlers. Each of these patches reduces the CPU
utilized per RPC by Send completion by an average of 2-3%.

The goal is to improve the rate of RPCs that can be retired for a
single-transport workload, thus increasing the server's scalability.

These patches are also available for testing:

https://git.kernel.org/pub/scm/linux/kernel/git/cel/linux.git/log/?h=for-next

---

Chuck Lever (3):
      svcrdma: Fewer calls to wake_up() in Send completion handler
      svcrdma: Relieve contention on sc_send_lock.
      svcrdma: Convert rdma->sc_rw_ctxts to llist


 include/linux/sunrpc/svc_rdma.h          |  7 +--
 net/sunrpc/xprtrdma/svc_rdma_rw.c        | 56 ++++++++++++++++--------
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    | 41 +++++++++--------
 net/sunrpc/xprtrdma/svc_rdma_transport.c |  4 +-
 4 files changed, 66 insertions(+), 42 deletions(-)

--
Chuck Lever

Comments

Tom Talpey July 26, 2021, 4:52 p.m. UTC | #1
On 7/26/2021 10:46 AM, Chuck Lever wrote:
> The following series improves the efficiency of NFSD's Send
> completion processing by removing costly operations from the svcrdma
> Send completion handlers. Each of these patches reduces the CPU
> utilized per RPC by Send completion by an average of 2-3%.

Nice gain. Are the improvements additive, i.e. 5-10% CPU reduction
for all three together?

Tom
Chuck Lever III July 26, 2021, 5:47 p.m. UTC | #2
> On Jul 26, 2021, at 12:52 PM, Tom Talpey <tom@talpey.com> wrote:
> 
> On 7/26/2021 10:46 AM, Chuck Lever wrote:
>> The following series improves the efficiency of NFSD's Send
>> completion processing by removing costly operations from the svcrdma
>> Send completion handlers. Each of these patches reduces the CPU
>> utilized per RPC by Send completion by an average of 2-3%.
> 
> Nice gain. Are the improvements additive, i.e. 5-10% CPU reduction
> for all three together?

Yes, the improvements are additive.

--
Chuck Lever