Re: RAID Performance
From: Andreas Schulze (b79xan_at_gmx.de)
Date: 03/25/04
- Next message: spamblanket: "AIX, TSM, Linux redbooks on ebay ...."
- Previous message: Jason Mather: "Re: Something like Linux NBD, but for AIX - ?"
- In reply to: Jim: "Re: RAID Performance"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Thu, 25 Mar 2004 17:06:31 +0100
"Jim" <jsmith@fluidmaster.com> schrieb im Newsbeitrag
news:82e9a4d8.0403190921.64e18258@posting.google.com...
> Thanks for the response Andreas. Here's a little more info...
>
> Our array is RAID5.
> Most of our access is random access on large files that are mostly
> included within one LV on the RAID. It accounts for about 90% of the
> RAID activity. All of the lvs for this database reside in this RAID.
>
> The fw cache is active.
>
> One 6230 with the blue cables.
>
> Here's typical output from iostat when the system is busy:
>
> tty: tin tout avg-cpu: % user % sys % idle %
> iowait
> 0.3 217.4 4.3 16.3 24.5
> 54.9
>
> Disks: % tm_act Kbps tps Kb_read Kb_wrtn
> hdisk0 59.9 279.9 67.5 2050 751
> hdisk7 21.0 81.4 18.7 64 751
> hdisk6 99.9 7582.7 451.0 74460 1424
> hdisk4 0.0 0.0 0.0 0 0
> hdisk1 0.0 0.0 0.0 0 0
> hdisk3 0.0 0.0 0.0 0 0
> hdisk5 0.0 0.0 0.0 0 0
> hdisk2 0.0 0.0 0.0 0 0
> cd0 0.0 0.0 0.0 0 0
>
> Lots of reads...not nearly as many writes. % disk utilization is very
> high as is I/O wait times.
>
>
Hallo Jim,
I agree with what Marc and KB suggested before: I also think you are best
off by redesigning your RAID set (and putting the lv in question into a
separate vg).
The basic problem might be that you are using a RAID 5 with just four
hdisks. Even if these four disks are available for the RAID itself (and not
for hot spare) you will hardly get good write performance with such a small
number of disks. SSA RAID5 has the lowest I/O request rate for all types of
writes compared to the RAID types 0,1 and 0+1. Personally I do not like SSA
RAID5 sets with less than seven disks as even with fw-cache enabled the
write performance is most of the time lousy (speaking of databases using the
RAIDs). You are right, the output suggests there are much more reads than
writes but the few writes account for most of the busy time of the disks. To
put it the other way round: there could be even more reads if the writes
would not throttle the system.
IMHO I guess your systems performance would profit from sth. like a four
disk 0+1 RAID without reducing availability much. Compared to a RAID5 I
would expect the write rate to almost double without/before tuning stripe
size etc. while the read rate will not suffer from that measure. The
'Advanced SerialRAID Plus Adapter Planning Guide' might be the adequate
source for preparing a decision on whether you want to stick with RAID5 or
to switch:
http://www.storage.ibm.com/hardsoft/products/ssa/docs/pdfs/ssaplan.pdf
HTH,
Andreas
- Next message: spamblanket: "AIX, TSM, Linux redbooks on ebay ...."
- Previous message: Jason Mather: "Re: Something like Linux NBD, but for AIX - ?"
- In reply to: Jim: "Re: RAID Performance"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
