Re: Oracle Performance on Sunfire T2000

From: Jeffery Thomas <>
Date: Mon, 16 Feb 2009 21:37:42 -0500
Message-ID: <>

We have two non-production 5120 and 5240 RAC clusters. We've noticed some single-thread
applications based on log miner with latencies in seconds on these boxes as opposed to milliseconds
on a 3-node v1280 cluster. Going by v$rman_backup_job_details, the RMAN backups to disk are
significantly faster on the v1280 than the 5120 for equivalently sized databases.

We have another two node v1280 production cluster and the business wants to expand this cluster
by adding a 5240 to it. Opinions? Not to mention that the v1280s NICs are ce-based while the 5240
is nxge, with SUN claiming we can get around this by creating soft links to build ce logical devices
pointing to the nxge physical devices.


On Thu, Feb 12, 2009 at 1:49 PM, Dennis Williams <> wrote:
> Pradeep,
> I think the 5220 & 5240 use the T2 chip. As I recall, the T1 chip had
> limited floating point capability. I think all the virtual CPUs shared the
> same FPU. I believe the T2 chip relives that particular problem.
> As stated in the note, the T1 and T2 chips offer lower single-thread
> performance, but a lot of multi-thread performance. For some Oracle
> applications, that works well, but if you have some critical single-thread
> processes, then you may see much lower performance. Sun offers a CoolThreads
> test that can run on your current Solaris server to give you an idea of
> whether your application will run well on a T2 server.
> Another issue is how the CPUs are counted. Oracle will usually get that
> number from the O.S., and Solaris on the T2 chip will tend to return the
> number of virtual CPUs. Sometimes this can cause you problems. If you create
> an instance and Oracle thinks there are 128 CPUs, it may allocate more
> memory than you expect.
> Dennis Williams

Received on Mon Feb 16 2009 - 20:37:42 CST

Original text of this message