Hello,

Using Sybase 12.5 on P630, AIX 5.2 with 4GB of RAM, of which 2.5GB is being
used by Sybase.

When running database checks on a Sybase database (it's checking page
allocation, etc), Sybase reports high disk I/O while the CPU isnt that busy.
See below:

Engine Busy Utilization CPU Busy I/O Busy Idle
------------------------ -------- -------- --------
Engine 0 4.0 % 96.0 % 0.0 %
Engine 1 2.0 % 98.0 % 0.0 %
------------------------ -------- -------- --------
Summary Total 6.0 % 194.0 % 0.0 %

When I run iostat it shows:

Disks: % tm_act Kbps tps Kb_read Kb_wrtn
hdisk2 3.5 28.0 7.5 0 56
hdisk5 100.0 2324.5 170.0 1472 3177
hdisk4 100.0 2388.5 174.0 1600 3177
hdisk3 3.5 28.0 7.5 0 56

All other disks have zero for their values. Now the disks with the high I/O
are where the logical volumes for the Sybase data devices are held, so
they're being hammered by the database checks. We've an SSA with 18 disks,
of which the Sybase installation, data and log devices are installed. I've
kept all Sybase stuff off the rootvg and on a 'datavg' volume group. All the
logical volumes on the 'datavg' volume group are mirrored (2 disks mirrored
to 2 other disks, so 4 in all). On an older Sybase 11.5 system, the AIX
sysadmin used disk striping. I decided against striping cause I was
uncertain our new sysadmin could set-up a RAID5 system. Is this a case of me
not using striping or are there obvious settings that is being overlooked.
This is a new system due to go in and the high I/O is worrying me.

TIA,
Martin

Martin Hanna wrote:
> Hello,
>
> Using Sybase 12.5 on P630, AIX 5.2 with 4GB of RAM, of which 2.5GB is being
> used by Sybase.
>
> When running database checks on a Sybase database (it's checking page
> allocation, etc), Sybase reports high disk I/O while the CPU isnt that busy.
> See below:
>
> Engine Busy Utilization CPU Busy I/O Busy Idle
> ------------------------ -------- -------- --------
> Engine 0 4.0 % 96.0 % 0.0 %
> Engine 1 2.0 % 98.0 % 0.0 %
> ------------------------ -------- -------- --------
> Summary Total 6.0 % 194.0 % 0.0 %
>
> When I run iostat it shows:
>
> Disks: % tm_act Kbps tps Kb_read Kb_wrtn
> hdisk2 3.5 28.0 7.5 0 56
> hdisk5 100.0 2324.5 170.0 1472 3177
> hdisk4 100.0 2388.5 174.0 1600 3177
> hdisk3 3.5 28.0 7.5 0 56
>
> All other disks have zero for their values. Now the disks with the high I/O
> are where the logical volumes for the Sybase data devices are held, so
> they're being hammered by the database checks. We've an SSA with 18 disks,
> of which the Sybase installation, data and log devices are installed. I've
> kept all Sybase stuff off the rootvg and on a 'datavg' volume group. All the
> logical volumes on the 'datavg' volume group are mirrored (2 disks mirrored
> to 2 other disks, so 4 in all). On an older Sybase 11.5 system, the AIX
> sysadmin used disk striping. I decided against striping cause I was
> uncertain our new sysadmin could set-up a RAID5 system. Is this a case of me
> not using striping or are there obvious settings that is being overlooked.
> This is a new system due to go in and the high I/O is worrying me.
>
> TIA,
> Martin
>
>
>
>
>
Well, that really isn't a lot to go on, but in all probability the
answer is 'yes'.

Without knowing the topology of your system (are you really talking
about 3 internal 6packs or do you mean you have 16 drives in a tower or
rack and a couple of SCSI drives in the system? How many cards? Are
you even using more than one loop?}, it is safe to say that spreading
the I/O over many disks is preferable to having one disk and it's mirror
heavilay banged on.

And frankly, you seem to use 'striping' and 'RAID5' as if they were the
same thing. There are similarities, but they are not the same.

Can't do much in the way of even general recommendations without knowing
what you have...if you have multiple SSA cards then striping/mirroring
would in _general_ be the fastest, if you only have one, may as well RAID5.