Hi,
I have some test on 10Gbe against NFS-based storage recently, and found the throughput of writing was not constant, the performance of writing drop down periodically.
My nfs client(version 3 and version 4 all tried) resident CentOS 6.6(2.6.32-504.el6.x86_64), NFS server is OpenMediaVault(5 ssd with stripe)
On a single mount point(or single nfs client), the avg throughput only can reach 430 MB/s, and two nfs clients aggregation can reach 700 MB/s, I can see there is periodical drop of network traffic graph both client and server side, and I am sure it's a nfs issue by lots of other tests.
I also found the nfs client commit procedure happened at the time that writing performance also dropped, and I have read B7 section(http://nfs.sourceforge.net) and think the relationship between them, but no many ideas, can someone help me pinpoint the root cause of this issue?
Can be related to flushing of dirty pages.
How much memory do you have on the clients?
You can play with vm.dirty_background_bytes
kernel parameter to get better performance.
Linux default is 10% of the RAM
(vm.dirty_background_ratio ), which can produce
spikes on IO if too much data accumulated in
fs cache, while you, probably, smooth continues
writing.
Tigran.
----- Original Message -----
> From: "Fu, Yong" <[email protected]>
> To: [email protected]
> Sent: Friday, September 9, 2016 7:34:54 AM
> Subject: NFS write throughput not constant
> Hi,
> I have some test on 10Gbe against NFS-based storage recently, and found the
> throughput of writing was not constant, the performance of writing drop down
> periodically.
> My nfs client(version 3 and version 4 all tried) resident CentOS
> 6.6(2.6.32-504.el6.x86_64), NFS server is OpenMediaVault(5 ssd with stripe)
>
> On a single mount point(or single nfs client), the avg throughput only can reach
> 430 MB/s, and two nfs clients aggregation can reach 700 MB/s, I can see there
> is periodical drop of network traffic graph both client and server side, and I
> am sure it's a nfs issue by lots of other tests.
>
> I also found the nfs client commit procedure happened at the time that writing
> performance also dropped, and I have read B7
> section(http://nfs.sourceforge.net) and think the relationship between them,
> but no many ideas, can someone help me pinpoint the root cause of this issue?
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html