Re: UltraSPARC IV

From: Michael Thomas <mhthomas_at_yahoo.com>
Date: Sun, 6 Jun 2004 04:41:30 -0700 (PDT)
Message-ID: <20040606114130.24458.qmail@web50607.mail.yahoo.com>

Hi,

Paul Drake <discgolfdba_at_yahoo.com> wrote:
> --- Michael Thomas <mhthomas_at_yahoo.com> wrote:
> > This can be amazing technology, especially in the
> > wrong hands. ;-)
>
> if you read the output of dmesg off of a linux box
> with a kernel that knows the difference between a
> physical and logical cpu, you'll see that the
> runqueue
> of logical is mapped to the physical. from just
> watching top, in a dual cpu system, the 2 physical
> CPUs will be loaded prior to any processes running
> on
> the logical(s). one can see overall cpu usage higher
> than 50% with hyperthreading enabled.
>

I have not played with this feature much. It really strikes me as something that can be easily mis-understood, so I posted my few experiences. Thanks for your detailed explanation on a linux system.

I've quit *killing myself* on bleeding edge new features to allow others the opportunity to *try getting the cheese* first.

OT (hijack): I've also enjoyed (sick) your AMD64 experiences. Since your recent postings I noticed a reviewer's article on AMD64 and Oracle was also limited to a *local* SCSI HD system. The author pointed out a storage limitation similar to your problem, but the vendor was less *truthful* (where's my tactful vocabulary this morning). Sorry I lost the link, but you probably already saw it, especially given your vendor. It would not help your evaluation anyway. ;-)

My last attempt to *force* a linux system to support new storage I/O interfaces was with SATA drivers, over a year ago. SATA is another marketing wonder. (/hijack)

> if the storage system is some form of RAID-F, the
> cpu
> wait % will be high enough that its not going to
> matter if you enabled or disabled hypterthreading
> anyways.
>

I was thinking along the same lines.

*Legally*, if you can prove that a hyperthreading Oracle server can *never* utilize this type of parallelization then why would someone pay for double the CPUs?

Does the dba need to run seti_at_home just to keep the CPU utilization high, along the lines of BCHR at 100%?
Many systems with hypertheading allow you to turn off the feature in BIOS, or possibly elsewhere, and avoid the *pay double licenses* penalty.

> Mike - I'm sure that I'm not saying anything that
> you
> don't already know.

Your info was news to me. Thanks again.

>
> Paul

Regards,

Mike Thomas

Do you Yahoo!?
Friends. Fun. Try the all-new Yahoo! Messenger. http://messenger.yahoo.com/

Please see the official ORACLE-L FAQ: http://www.orafaq.com

To unsubscribe send email to: oracle-l-request_at_freelists.org put 'unsubscribe' in the subject line.

--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Received on Sun Jun 06 2004 - 06:38:17 CDT