Re: Performance Question

Gavin

Do you have the stats on DB_Block_size for both databases, the number of db_block_buffers (whether LRU latches have been set), what patch level Oracle is at (Oracle lists several performance improvements in patches for 8.1.7), what the number of Block buffer reads for the SQL are (both Dell and Sun) and the timing.

As you have said no physical reads the disks won't cause an issue with this problem.

Have you ensured there are no rogue processes on the Sun box (top, if installed, or prstat - Solaris 8) may be useful. I assume the data (and meta-data) is identical between the 2 servers.

If a very large number of data blocks are being scanned you might look at setting up some LRU's for the buffer cache.

Xeon (I assume pentium 3 based - as speed is 700MHz) runs at 100Mhz (unsure what CAS level you may have and whether ECC or not). Sparc 3 systems will have memory either running at 150Mhz (backplane speed for Sparc 3 processors) are a fraction of that maybe 50 or 75 MHz - I assume memory is ECC and again don't know the CAS level.

If Sun is ECC and Dell isn't that could account for a speed difference in Dell's favour. If Dell memory has a lower CAS that would also give it a memory advantage.

Hope this helps

Nathan
"gdas" <gdas1_at_yahoo.com> wrote in message news:7a4ed455.0203301819.45eadc67_at_posting.google.com...
> Hi,
>
> I recently moved a database running Oracle 8.1.6 from Windows 2000 to
> Oracle 8.1.7 on Solaris 8. This was part of a hardware upgrade that
> was intended to improve performance and scaleability. I'm currently
> running both systems for performance comparisions right now.
>
> The Win2K hardware was a Dell Poweredge Server, Dual Xeon Processor
> 700 mhz with 1 GB of RAM. Disk subsystems consisted of 2 internal
> drive sets, 1 RAID-1 mirror for
> OS/Oracle.exe/Redo/Rollback/Temp/System and another drive set
> consisting of 5 disks configured in RAID-5 fashion for datafiles.
>
> The new hardware is a SunFire 280R, Dual 900 mhz with 2 GB of RAM.
> Disk subsystems consist of 1 internal drive set that is RAID 1 for OS,
> oracle.exe, redo, rollback, system, temp. 2nd drive set is an
> external high performance fibre channel disk array with 12 disks that
> are configured in RAID10 (stripped and mirrored). this contains the
> oracle datafiles.
>
> After exporting/importing the data, rebuilding the indexes and
> analyzing everything, first observation was that the new disk array is
> VERY FAST.
>
> Because the new sun box has twice the RAM, I doubled the size of the
> db_block_buffers and the shared pool size as well.
>
> I'm not a unix expert. (I am working alonside a unix sysadmin who
> setup the machine). by the way, he did properly setup the parameters
> in the /etc/system according to oracle's instructions.
>
> What I am noticing however is that memory reads/buffer gets seem to be
> slow. (slower than the Dell)
>
> I run a query twice such that it's in the SGA. The trace on the query
> shows 0 physical reads and all work in "buffer gets"... It is
> significantly slower on the sun box.
>
> I've checked iostat and vmstat and there is no swapping happening.
>
> Is there any kind of tuning or oracle level parameter I might want to
> look at or something else to investigate (please bear with my lack of
> unix knowledge)?
> Is there anyway to improve the performance of memory reads? Something
> in oracle? Something at the OS Level?
>
> or is this simply the way it is? That Solaris RAM is not as fast and
> Dell RAM? I hope that's not the case...I hope we didn't make the wrong
> Hardware decision.
>
> Thanks for any advice.
> Gavin
Received on Sun Mar 31 2002 - 00:28:17 CST