upgrading server
Fairlight
fairlite at fairlite.com
Mon Mar 28 12:16:03 PST 2005
Is it just me, or did Lerebours, Jose say:
> Enrique posted:
>
> > Hi,
> >
> > Anybody here running a Proliant ML-370 or ML-570 server with filepro
> > and linux ? The ML-370 has 2 Xeon 3.6Ghz/800FSB/2MB cache and the
> > ML-570 has 4 of them (CPU's) same specs. Will we notice a difference
> > with filepro or it doesn't matter because the Hard drives speed ( 15K
> > SCSI) set the pace for filepro database processing ?.
The drive speed sets part of the pace for disk I/O operations. The
bus sets the other part of that pace. For actual -processing- that is
unrelated to disk I/O, the CPU's, RAM, bus speed, and cache all come into
play.
> > We're in the process of upgrading our 3 year old P3 1.4Ghz server and
> > maybe OS to linux and I want to see if somebody has any input of these
> > servers with RED HAT and filepro/vsifax
The only input I have is that you will notice speed increases, but not
necessarily in the proportions you might assume. You're going to faster
chips, presumably a faster FSB, and I'd wager bigger L1 cache as well.
That will give you a performance boost on a single-chip system in the
proportions you'd expect.
However, going to a quad-CPU will not necessarily give you 4x the
performance--probably not even close. filePro is non-threaded. That is,
each process runs as a single thread in a single process. Since only
one CPU is used for a single thread (or single-threaded process), if you're
running only one at a time, you're only going to see the improvement you'd
see on one chip. However, you have three other chips there, so you can
handle more processes between them with that much more ease.
Of course, "that much" is a relative term, as you still have contention for
memory and disk I/O. You won't even get a linear improvement by adding more
chips, much less exponential or logorithmic. Past a certain point, the
overhead of doing SMP management also starts to take part of the
processing. This is accounted for in something like that 64000 processor
system that one university just put together, but by and large, I've heard
that going past 8 is non-productive in many scenarios.
It really depends on the entire architecture, and resource contention
versus thread/process count.
> Am I getting this right or it seem that your actual intend is to joke
> around ... Are you really concerned on downgrade performance if you
> upgrade from P3 1.4Ghz to "an actual" server with 2 or 4 Xeon 3.6Ghz with
> 2MB RAM?
He specifically mentioned HD's.
> If filePro chokes on a dual/quad xeon processor server, dump it! No, not
> the server, filePro.
Jose, that was one of the less intelligent comments you've made here.
There's nothing inherent about to an SMP system that will cause filePro to
fail. Nor is there any limit in filePro that would preclude using it on an
SMP system. The only trouble I've ever seen had to do with a 64-bit
platform with non-stop-clustering (UnixWare 7.1.1 with NSC, specifically)
where the PID space actually -started- above the normal threshhold for
most 32bit systems, thus causing fork() to think it always failed--which
was indeed an erroneous assumption. A recompile with the correct devkit
fixes things like that. But that wasn't even an SMP issue--and NSC has
long since vanished from the UW offerings.
To make a comment to the effect you did about filePro on SMP systems is
an egregious display of poor judgement.
> Nice to learn that filePro is more dependent on HD speed that it is on
> RAM - I did not know this? I guess that this should be true for some
Depends what you do more of--disk I/O or actual opcode processing.
> We have an old proliant 7000 with 18 HDs & 4 Xeon 500Mhz and 1.0G RAM
> with an average load of 110 users and the CPUs barely blink. When we
> upgraded from a Quad 200Mhz, we found that the speed went up drastically.
If you already have evidence that it's not a problem, why make the comment
you did above? You're making less sense by the second, sir.
> I would not expect you to have a decreased performance.
Nor would I. Quite the contrary. I've seen fP on dual and quad Xeon 2.4's
and 2.8's. Actually, -everything- runs lightning fast on those things.
Especially the Xeons, if you have hyperthreading enabled, actually read out
as if they were dual core chips. They're not, but they simulate it in a
way with which I've never bothered to become fully conversant. If you have
four physical chips, hyperthreading will make it appear as if you have eight.
I have to really question how much someone is doing with fP and VSI*FAX to
require a quad Xeon 3.6 system, however. VSI is going to be limited by the
size of your modem pool. Just how many users do you have on your license,
for example? Past a certain point, people seem to just toss money at
machines, believing that bigger is better. It is--when it doesn't go to
waste. And I've seen some fP systems with a LOT of wasted CPU. Having 94%+
of all four CPU's idle 99% of the time is a real indicator that you don't
really need that huge system, for example.
It pays to do an analysis with sar (or at least a decent study using top at
regular intervals) before shooting for an upgrade because you think you
need more power. In many instances, you might be upgrading the wrong
subsystem at great expense. Xeons don't grow on trees.
mark->
--
***** Fairlight Consulting's Software Solutions *****
OneGate Universal CGI Gateway: http://onegate.fairlite.com/
FairPay PayPal Integration Kit: http://fairpay.fairlite.com/
RawQuery B2B HTTP[S] Client & CGI Debugger: http://rawquery.fairlite.com/
Lightmail Mail Sending Agent: http://lightmail.fairlite.com/
More information about the Filepro-list
mailing list