Re: vfs.nfsrv.async broken ? Or pilot error ?

From: Brian Reichert <reichert_at_numachi.com>
Date: Fri, 28 Jan 2005 17:44:12 -0500
On Fri, Jan 28, 2005 at 05:08:17PM -0500, Mike Tancsa wrote:
> I am trying to run some nfs benchmarks and wanted to see how 
> vfs.nfsrv.async makes a difference on performance for FreeBSD vs Linux.
> 
> From the client (a RELENG_5 box) if I do a simple
> 
> time dd if=/dev/zero of=/mnt/testfile3 bs=16k count=1638
> 
> The write is very quick
> 
> [ns4-new]# time dd if=/dev/zero of=/mnt/testfile3 bs=16k count=1638
> 1638+0 records in
> 1638+0 records out
> 26836992 bytes transferred in 0.387338 secs (69285733 bytes/sec)
> 0.007u 0.079s 0:00.54 12.9%     34+266k 0+3276io 0pf+0w
> [ns4-new]#
> 
> 
> 
> But if I increase the size by 10 times, the process takes more than 100 
> times the amount of time to complete, and sometimes it doesnt complete at 
> all

I've been trying to diagnose some big-files-over-NFS issues myself,
with Solaris clients and a 5.3-R server.  

The symptoms on my setup were: when the files got to 200M, the nfsd
processes were seemingly hang in a getblk state, and, after many
(tens) of minutes, would 'wake up', and continue to flow data.

My current workaround was to disable softupdates on the exported
partitions until I can get another maintenace window to do things
like check the firmware of my RAID card. etc.   There are probably
dozens of important differences between your and my systems, so I'm
not going to pretend we have the same problems, but that's what
I've done for now...


-- 
Brian Reichert				<reichert_at_numachi.com>
37 Crystal Ave. #303			Daytime number: (603) 434-6842
Derry NH 03038-1713 USA			BSD admin/developer at large	
Received on Fri Jan 28 2005 - 21:44:15 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:27 UTC