Skip to contentSkip to conversation

svcrdma: Increase the server's default RPC/RDMA credit grant

PATCH Superseded Chuck Lever 2 participants 4 messages Sep 26, 2025 – Oct 1, 2025

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Chuck Lever

author maintainer
#0 lore raw

PATCH 1/1 svcrdma: Increase the server's default RPC/RDMA credit grant

From: Chuck Lever <[email protected]>

Now that the nfsd thread count can scale to more threads, permit individual clients to make more use of those threads. Increase the RPC/RDMA per-connection credit grant from 64 to 128 -- same as the Linux NFS client.

Simple single client fio-based benchmarking so far shows only improvement, no regression.

---
 include/linux/sunrpc/svc_rdma.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/sunrpc/svc_rdma.h +1 -1
index 22704c2e5b9b..57f4fd94166a 100644
@@ -131,7 +131,7 @@ static inline struct svcxprt_rdma *svc_rdma_rqst_rdma(struct svc_rqst *rqstp)
  */
 enum {
 	RPCRDMA_LISTEN_BACKLOG	= 10,
-	RPCRDMA_MAX_REQUESTS	= 64,
+	RPCRDMA_MAX_REQUESTS	= 128,
 	RPCRDMA_MAX_BC_REQUESTS	= 2,
 };
 
Signature · 1 line
2.51.0
#1 lore raw
Quoted 3 lines · Chuck Lever

From: Chuck Lever <[email protected]>

Now that the nfsd thread count can scale to more threads, permit

Just trying to appreciate which change(s) paved the way for this RPCRDMA_MAX_REQUESTS change.

Are you referring to the netlink interface changes Jeff did earlier this year or something else? (thinking "something else" but...)

Might be useful to update the header to convey which specific commit(s) made this change possible.

Mike

Quoted 29 lines

individual clients to make more use of those threads. Increase the RPC/RDMA per-connection credit grant from 64 to 128 -- same as the Linux NFS client.

Simple single client fio-based benchmarking so far shows only improvement, no regression.

---
 include/linux/sunrpc/svc_rdma.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/sunrpc/svc_rdma.h +1 -1
index 22704c2e5b9b..57f4fd94166a 100644
@@ -131,7 +131,7 @@ static inline struct svcxprt_rdma *svc_rdma_rqst_rdma(struct svc_rqst *rqstp)
  */
 enum {
 	RPCRDMA_LISTEN_BACKLOG	= 10,
-	RPCRDMA_MAX_REQUESTS	= 64,
+	RPCRDMA_MAX_REQUESTS	= 128,
 	RPCRDMA_MAX_BC_REQUESTS	= 2,
 };
 
Signature · 1 line
2.51.0
author maintainer
#2 lore raw
Quoted 5 lines · Mike Snitzer
Quoted 3 lines · Chuck Lever

From: Chuck Lever <[email protected]>

Now that the nfsd thread count can scale to more threads, permit

Quoted 8 lines

Just trying to appreciate which change(s) paved the way for this RPCRDMA_MAX_REQUESTS change.

Are you referring to the netlink interface changes Jeff did earlier this year or something else? (thinking "something else" but...)

Might be useful to update the header to convey which specific commit(s) made this change possible.

The svc thread scaling change refers to the commits from

e3274026e2ec ("SUNRPC: move all of xprt handling into svc_xprt_handle()")

to

15d39883ee7d ("SUNRPC: change the back-channel queue to lwq")

all dated about two years ago, merged in v6.7. Just checking, should the updated description provide more detail than that?

Signature · 1 line
Chuck Lever
#3 lore raw
Quoted 17 lines · Chuck Lever
Quoted 5 lines · Mike Snitzer
Quoted 3 lines · Chuck Lever

From: Chuck Lever <[email protected]>

Now that the nfsd thread count can scale to more threads, permit

Quoted 8 lines

Just trying to appreciate which change(s) paved the way for this RPCRDMA_MAX_REQUESTS change.

Are you referring to the netlink interface changes Jeff did earlier this year or something else? (thinking "something else" but...)

Might be useful to update the header to convey which specific commit(s) made this change possible.

The svc thread scaling change refers to the commits from

Quoted 8 lines

e3274026e2ec ("SUNRPC: move all of xprt handling into svc_xprt_handle()")

to

15d39883ee7d ("SUNRPC: change the back-channel queue to lwq")

all dated about two years ago, merged in v6.7. Just checking, should the updated description provide more detail than that?

No, that's perfect. Thanks