xfs
[Top] [All Lists]

Re: Playing around with NFS+XFS

To: Federico Sevilla III <jijo@xxxxxxxxxxxxxxxxxxxx>
Subject: Re: Playing around with NFS+XFS
From: Seth Mos <knuffie@xxxxxxxxx>
Date: Thu, 30 Aug 2001 22:37:01 +0200 (CEST)
Cc: "Philippine Linux Users' Group Mailing List" <plug@xxxxxxxxxxxxxxxxx>, Linux XFS Mailing List <linux-xfs@xxxxxxxxxxx>
In-reply-to: <Pine.LNX.4.33.0108310247460.960-100000@gusi.leathercollection.ph>
Sender: owner-linux-xfs@xxxxxxxxxxx
On Fri, 31 Aug 2001, Federico Sevilla III wrote:

> Hi everyone,
> 
> The mount options for NFS are "rsize=8192,wsize=8192". I think I can push
> things further by bumping this up to 32768 since I'm using NFSv3, but
> maybe I should sleep first. ;>

I use 16384 for 100Mbit at work which seems to be a decent size vs
reponse ratio.

> Anyway here are the results:
> 
> Version  1.01d      ------Sequential Output------ --Sequential Input- 
> --Random-
>                     -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- 
> --Seeks--
> Machine        Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
> Gusi-NFS-XFS     1G  4200  82  5937   7  1988   4  4037  76 10512   9 120.3   
> 1
>                     ------Sequential Create------ --------Random 
> Create--------
>                     -Create-- --Read--- -Delete-- -Create-- --Read--- 
> -Delete--
> files:max:min        /sec %CP  /sec %CP  /sec %CP  /sec %CP  /sec %CP /sec %CP
> Gusi-NFS-XFS  5:1:0    80   2   252   5    56   1   118   4   341   7 41   0
> 
> A more readable copy is in <http://jijo.leathercollection.ph/bonnie.html>.
> 
> I'm a definite newbie when it comes to this benchmarking business. I
> prefer to read everybody else's statistics. Maybe someone has some other
> decent stress tests to recommend?

I have a few standard bonnie results of linux -> linux tests on my
homepage http://iserv.nl/ 
They are a few entries down and is in the following config.

Server was a pIII 450 with 256MB of ram and a 2c905B NIC and a
40GB IDE disk in UDMA33 mode. I used 8 nfsd processes and enlarged the
buffer size from 64KB to 256KB.
echo 262144 > /proc/sys/net/core/rmem_default
echo 262144 > /proc/sys/net/core/rmem_max

Client was a Dual PIII 733 with 256MB ram and a Intel eepro100 NIC.

The limiting factor was wire speed though.

You may note that the speeds I was clocking on this desktop machine with 1
IDE harddisk in UDMA33 mode on a i440BX motherboard are higher then your
scores for the server.

However it seems Linux and XFS is a fast combo for large databases. We
needed to do a dictionary update of our databases (6GB total) and since we
have a testmachine (the PIII450 Desktop model from above) and Progress 9
we could simulate the entire process.

The production server is a dual PII400 with 512MB of ram and 45GB of disk
in raid5 under NCR MP-RAS (SVR4) and Veritas FS.

The conversion was done on the sunday when there was nobody in the office
and the machine was totally idle. The testmachine was during a workday and
about 2 people were busy upsetting the printing system (including me
Yay!). 

The conversion under linux took 35 minutes and the machine was responsive.
The conversion under NCR MP-RAS took 6 Hours (!) and the machine was
stuttering along :-(

My manager urges me to make the migration from the NCR to Linux machine on
a faster schedule. No more convincing left to do. The new machine will
be much faster then the current desktop machine so he also took this
oportunity to take a vacation.

Hope you get some more info.

Cheers
Seth


<Prev in Thread] Current Thread [Next in Thread>