Date: Fri, 31 Jan 2014 20:41:05 -0500 From: J David <j.david.lists@gmail.com> To: Rick Macklem <rmacklem@uoguelph.ca> Cc: freebsd-net@freebsd.org, Garrett Wollman <wollman@freebsd.org> Subject: Re: Terrible NFS performance under 9.2-RELEASE? Message-ID: <CABXB=RSCHTg662Fj2CBZHD=8-GEaZyYdVRcsCTxBbaJ-f8ZwuA@mail.gmail.com> In-Reply-To: <1609454808.1083115.1391210456671.JavaMail.root@uoguelph.ca> References: <CABXB=RQ6LdeoNi4vNZGCaM2C_up_JCf2SpWPzm2S_M_%2BpTnzsQ@mail.gmail.com> <1609454808.1083115.1391210456671.JavaMail.root@uoguelph.ca>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jan 31, 2014 at 6:20 PM, Rick Macklem <rmacklem@uoguelph.ca> wrote:
> Oh, and remember to try setting readahead=8 in your mounts, too. NFS will
> do a read + N readaheads (where N == 1 by default) and then wait for
> replies to those before continuing on.
Predictably, this has no effect on anything but sequential reads.
No tuning is going to change the fact that writing 14MiB/sec from the
client to the server results in 200+ MiB/sec of wasted traffic being
sent from the server back to the client.
This is from the client's interface during a write-only test:
Interface Traffic Peak Total
vtnet1 in 202.838 MB/s 219.467 MB/s 359.898 GB
out 14.127 MB/s 14.346 MB/s 96.503 GB
If write performance did get to wire speed on this workload, the most
it could ever do would be <128MiB/sec, because the unused backflow of
2GiB/sec would max out the interface.
Thanks!
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CABXB=RSCHTg662Fj2CBZHD=8-GEaZyYdVRcsCTxBbaJ-f8ZwuA>