Re: Perfomance problem with LV mirroring

Or the new IBM ESS (shark) box which is quicker than the EMC box !

In considering your argument, normally the SSA architecture is not the bottleneck, it is normally down to the levels of memory caching that are done by the server itself. If the rdbms has to go off to the disk subsystem for a read (writes later!) then unless the memory of the host is limited (less than the disk subsystem) then it is unlikely that the data will be in the cache of the subsystem either, and will have to read from disk anyway.

On the issue of writes, normally this is where the ESS or EMC box, etc will have a performance benefit as the writes are cached in the memory and NVS of the disk subsystem and the write delay/penalty is greatly minimised.

For your particular example, I am assuming that you have you hardware set up with one loop mirrored to another loop across two adapters and both I/O bays in the J50.

Yes you can get issues with the performance of the redo log volumes, but this can be minimised, by

1. Only having the redo logs on the volumes
2. Correct size of the redo log volumes, if they are being cut very frequently then they should be increased in size.
3. The redo log disks should be in the middle of the loops so that the access to the disks is uniform, i.e same number of hops no matter which side of the loop (1 or 2) the I/O gets initiated from.

These are just a few of the basic things that can be done, if you need to know more or want to discuss a point (argue !), send me a e-mail.

Regards, John.

In article <809vd4$qin$1_at_news1.skynet.be>,   "GDN" <gert.deneve_at_advalvas.be> wrote:
> If you have a lot of I/O, the SSA architecture can become a
bottleneck. All
> the data from the SSA-disks have to come across the same wire. Also
you
> don't have any caching.
> For very busy rdbms it may be wise to consider another architecture. (
eg.
> EMC box ? I know it's very expensive )
> Real time example :
> v7.3.4 rdbms on AIX 4.2.1.0
> J50 cluster ( HACMP )
> about 480 users
> about 48 drives ( 2 Gb / 4.5 Gb ) in 2 locations
> using AIX mirroring ( lv level )
> all the files are on the SSA-disks
> -> bottleneck for the redo log writer
> We have a lot of redo wait times > 30 msec
>
> Hope this helps
> Gert
>
> John Newman wrote in message <807frb$kea$1_at_nnrp1.deja.com>...
> >No performance problems that I am aware of there are a couple of
> >important fixes for async I/O around, but that is all. As usual make
> >sure that you have go the async I/O tuning right for min and max
servers
> >
> >In article <806243$t02$1_at_porthos.nl.uu.net>,
> > "Wil \"Spock\" Weterings" <weterings_at_schiphol.nl> wrote:
> >> I heard of a rumour that there is a problem with AIX 4.3.x and LV
> >mirroring.
> >> I have an Oracle parallel server application with two RS/6000
> >(G40/F40) and
> >> a SSA disk unit that uses raw volumes (concurrent acces) and
mirroring
> >with
> >> no performace problem that i am aware of (Oracle 8.0.4. with about
50
> >> sessions).
> >>
> >> We are buikding an application that is must bigger over 200 users
and
> >H70
> >> machines and SSA disks and are using Oracle Apps. Is there someone
who
> >knows
> >> of perormace problems.
> >>
> >> Please respond to weterings_at_schiphol.nl.
> >>
> >>
> >
> >
> >Sent via Deja.com http://www.deja.com/
> >Before you buy.
>
>

Sent via Deja.com http://www.deja.com/
Before you buy. Received on Tue Nov 09 1999 - 16:38:58 CST