x4500 niagara

So if the T1 processor is so good at throughput, and one would expect
an NFS (or iSCSI) environment to be able to take very good advantage
of this, why is the x4500 an x64 box ... along with the attendant misery
of BIOS vs OBP/ALOM.

Not to mention power consumption.

-frank

 

Good question.

I think the power delta between one T1 and a pair of Opeterons would
be small compared to 48 SATA disks...

--
Rich Teer, SCNA, SCSA, OpenSolaris CAB member

President,
Rite Online Inc.

Voice: +1 (250) 979-1638
URL: http://www.rite-group.com/rich

 

Not really. Hard drives don't consume a lot of power. Also, Sun
is obligated to use the SWaP metric.

Only 11.5W for a single hard drive "random R/W operation" (from
http://www.pcguide.com/ref/hdd/perf/perf/spec/otherPower-c.html)
x48 = 552W

vs 72W for T1 power (13% of hard drive power)
and 95Wx2 = 190W for Opteron power (34% of hard drive power)

And the difference is 118W, but then factor in 4 threads vs 32 threads.
8x the throughput for 85% of the power and I have to ask again, why
is this not a T1 processor?

-frank

 

I think Thumper was originally from Andy B's Kaela (sp?) that Sun bought.
a year or two ago. If that's the case, it might simply be because Kaela
didn't have access to a T1 early enough in the design cycle.

But yeah, I'd be interested in seeing how the performance of a T1 based
X4500 would compare to the Opteron based one.

--
Rich Teer, SCNA, SCSA, OpenSolaris CAB member

President,
Rite Online Inc.

Voice: +1 (250) 979-1638
URL: http://www.rite-group.com/rich

 

The x4500 comes with the "ILOM" remote management (separate IP for the
ILOM and ssh access to the console port. impitool remote powermanagement,
etc. So it is not all that bad.

Oh, there's also the Java KVM which is builtin (run the VGA console from
within a Java application and even attach ISO images)

With 500Watt for the disks, who is counting CPU Watts?

Casper

 

So that's 18% power increase (less counting DIMMs)

The main reason, I think, is largely historic; the system was
far in its design stage when Sun bought the company designing it.

Casper

 

Any word on if newer x86 firmware will be smarter than generic
legacy BIOS? Even Apple is using EFI, and having something comparable to
OF (or actually using OF syntax, if not being actual OF) would be
nice.

 

That would be nice, but my very limited understanding of such things
tells me that the BIOS is required to support Windoze, a requirement
Apple presumably doesn't have. If the former is true, it would be
another example of braindead stuff relying on braindead stuff...

--
Rich Teer, SCNA, SCSA, OpenSolaris CAB member

President,
Rite Online Inc.

Voice: +1 (250) 979-1638
URL: http://www.rite-group.com/rich

 

Watts/work, not just watts.

-frank

 

Boot Camp for Apple will run Windows XP, and all Apple hardware
uses EFI instead of legacy BIOS.. I wouldn't be surprised if there were
some form of emulation involved to pass muster with XP, but it should be
possible.

Frankly, having to sit and wait through each peripheral's 'hit
ctrl-X in the next 5 seconds to configure RAID/PXE boot/go into
setup/flush the toilet' is extremely annoying and very 20th century.

 

How do you arrive at 8x the throughput? Somehow assuming 1 Opteron
"thread" == 1 T1 "thread" doesn't seem quite right.

rick jones

 

Where would you place it? Sun certainly is positioning it as pretty
high performance. (But I don't disagree, 8x is not going to be met.)

http://www.sun.com/servers/coolthreads/t1000/benchmarks.jsp#m

It's ashame they don't put it up against other Sun servers, THAT
would be good info.

-frank

 

* hennessy:

Right, just that one part of Boot Camp is a firmware update which
implements the missing BIOS part in the firmware of Apples intel Macs...

Neither Windows 32bit nor x64 support EFI. Same is valid for Vista btw.
The only Windows that supports EFI are the Windowsxp and Windows Server
2003 versions for IA64 Itanium...

Benjamin

 

A good question. Even if I had a concrete answer, I'm not sure I
could say since so much of the stuff one needs to run on the thing (at
least under Solaris) has an EULA that specifically forbids discussion
of evaluation/benchmarking without the express written consent of Sun.

More data is always good.

rick jones

 

Well, not if you're trying to make a case for war. ;-)

 

I know this is the case with S10 FCS, is it the same with S10 6/06?

If so, how does one legally blog their evals for JS and company's
Try-n-Buy program? Using a non-Sun OpenSolaris distribution?

John

 

It certainly looked that way. Studio 11 also.

I am not a laywer (the acryonym seemed too ripe for
mis-interpretiation). Indeed the JS blog entry was interesting. I'm
still trying to learn all the ramifications but for the time being
would have to assume that it was limited to those who were using
try-n-buy on the T[12]000.

rick jones