Re: 10gR2 RAC on RISC or X86-64 hardware
Date: Mon, 17 Mar 2008 17:43:33 -0700
I'm not a HW guy either, nor do I play one on TV.
But I do know some HW guys. The common nickname for Itanium is 'Itanic'
Make of it what you will.
The current IO champ in price/performance is not Intel, but AMD Opteron.
Don't know if it has been through TPC, but I do know some folks that
IO and CPU's down to the bit level, and their choice is Opteron. I believe HP has
a line of servers with this chip.
Very fast with Oracle.
On Mon, Mar 17, 2008 at 12:33 PM, Jeremy Schneider < jeremy.schneider_at_ardentperf.com> wrote:
> On Mon, Mar 17, 2008 at 8:34 AM, Joop Gijsbers <jg_dba_at_xs4all.nl> wrote:
> > I have a discussion with a colleague, Technical consultant with a
> > hardware/IBM/AIX background about the benefits of the use of IBM p- series
> > hardware instead of x86-64 hardware.
> > His these is that because of Oracle Licensing RISC based hardware -
> > especially the IBM hardware - always will be more profitable then a
> > Linux/ASM architecture on X86-64 hardware.
> I'm not an in-depth hardware guy either but I just went to the TPC
> website... and in the TPC-C (clustered and non-clustered) and in every
> single size category for the TPC-H, the system with the best published
> price/performance (which includes licensing) is either Itanium or x86[-64].
> I'm not saying this closes the case, but to so quickly dismiss the Intel
> platform is simply ignorant.
> From his experience he has a " rule of thumb" that there is a maximum of
> > 250 concurrent " users" on a RISC based CPU and a maximum of 100 concurrent
> > " users" on a Intel-based CPU. So because of the CPU licenses he argued
> > that Oracle on Intel-based CPU will mostly more expensive then on RISC-based
> > CPU, especially in the case of RAC (with EE).
> In my opinion those "rules of thumb" are completely arbitrary and totally
> useless. The number of users on a system completely depends on what you're
> doing. For a decision support or analytical system (I/O driven ad hoc
> queries) you might be lucky to get 10 concurrent users on a CPU. For a
> website (80% small reads on the cache) I think you'd get laughed at for only
> getting 250 concurrent users per CPU.
> Jeremy Schneider
> Chicago, IL
-- Jared Still Certifiable Oracle DBA and Part Time Perl Evangelist -- http://www.freelists.org/webpage/oracle-lReceived on Mon Mar 17 2008 - 19:43:33 CDT