Re: performance of san-disk

From: Holger.VanKoll@SWISSCOM.COM
Date: Wed Nov 13 2002 - 06:00:26 EST


I dont use jfs2. Anyway, I read directly from the disk in the script; so
the fs thats on the disk shouldnt matter.
 

        -----Original Message-----
        From: Darryl Ousterhout [mailto:D.Ousterhout@LABSAFETY.COM]
        Sent: Tuesday, November 12, 2002 10:55 PM
        To: aix-l@Princeton.EDU
        Subject: Re: performance of san-disk
        
        
        Not sure if this is your problem or not, but I was talking with
IBM last week about a known JFS2 performance problem in 5.1. If you are
using JFS2 on file systems other then rootvg, it has been known to cause
severe performance problems. Check to see if you have APAR IY32280
installed. This was IBM's fix for the JFS2 problem.
         
        Regards,
        Darryl
         
         
         -----Original Message-----
        From: Holger.VanKoll@SWISSCOM.COM
[mailto:Holger.VanKoll@SWISSCOM.COM]
        Sent: Tuesday, November 12, 2002 3:05 PM
        To: aix-l@Princeton.EDU
        Subject: performance of san-disk
        
        

                Hello,

                I got a p670 connected to HDS (Hitachi; 9900) Storage
with 2 FC-adapter.

                No matter how much disks I access concurrently, I get
34mbyte/sec maximum.
                All 4 CPU are 0% idle when measuring, but depending on
the # disks they spend more time in kernel than in wait.

                # disks accessed kernel wait disk-busy
(topas) transfer Mbyte/sec
                10-20 8 90 50-70%
25
                30 40 60 40-60%
33
                40 55 45 60-80%
33
                50 70 30 40-60%
34
                60 78 22 35-50%
34

                This looks like my CPU-power is the bottleneck, correct?

                As I get 184 Mbyte/sec with the 4 internal disks and 60%
kernel / 40% wait that looks to me as if the fibre-channel driver
(kernel-time) uses all my cpu. Is this correct?

                If someone likes to do some measurement to have numbers
to compare, here is the script I used:

                #!/bin/sh
                #dlmfdrv4 - 61
                count=2000
                bsk=256k
                dev=dlmfdrv
                #dev=hdisk

                start=4
                end=54

                i=$start
                until false
                do
                dd </dev/$dev$i >/dev/null bs=$bsk count=$count
2>/dev/null &
                [ $i = $end ] && break
                let i=i+1
                done

                I would appreciate any other "benchmark"-results,
especially latest ssa-technology and non hds (f.e. ibm) storage servers.

                Regards,

                Holger



This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 22:16:19 EDT