Date: Thu, 30 Jan 2020 23:10:16 +0000 From: Rick Macklem <rmacklem@uoguelph.ca> To: freebsd-net <freebsd-net@freebsd.org> Subject: NFS performance depends on low latency reception of small TCP segments Message-ID: <YTBPR01MB33741212B594153F97F0AD09DD040@YTBPR01MB3374.CANPRD01.PROD.OUTLOOK.COM>
next in thread | raw e-mail | index | archive | help
Hi,=0A=
=0A=
I have seen recent discussion related to NET_EPOCH.=0A=
This is way out of my expertise, but...=0A=
=0A=
NFS traffic is basically bi-directional small messages.=0A=
If a change increases the latency of reception of a small=0A=
message (TCP segment) which is not followed by further=0A=
TCP segments in the same direction...=0A=
--> A significant NFS performance hit could be observed.=0A=
=0A=
I don't have hardware that can do reasonable benchmarking,=0A=
so if some else can test for NFS performance regressions=0A=
when making TCP stack changes, it would be appreciated.=0A=
--> The time it takes to do a fairly large build like a kernel=0A=
or buildworld over NFS on fast hardware with fast networking=0A=
would probably do it.=0A=
=0A=
Just wanted the people doing TCP stack work to be aware of this, rick=0A=
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?YTBPR01MB33741212B594153F97F0AD09DD040>