Dual Xeon at Work - Need Help

jen4950

[H]F Junkie
Joined
Apr 25, 2001
Messages
11,391
Hello-

I've got a brand spanking new Dual Xeon with HT at work; i.e. I see 4 graphs at the 'pulse window' as I like to call it (task manager). It's easy to see if my model is still running :D

I use ETABS for stuctural analysis- and have an absolutely enormous model. a 35MB model in ETABS makes about 5GB in analysis files each run. And it's averaging about 8 hours per run.

I've noticed that it doesn't seem to be using both processors/ all 4 logical processors. At full tilt it's using 25% of CPU time.

Anyone have any ideas on how to tweak this badass multiprocessor system for what it was bought for? i.e. how do I get the 25% CPU usage up to 100% when using ETABS? Related- the whole goal here is to cut processing time for my huge model from 8 hours to 2 hours with the same machine and software.

https://www.csiberkeley.com/

As an example, the model is running right now- and it will probably finish at about 6:00 in the morning on Sunday. And any changes that you make require a new run. This isn't a problem for normal models that take about 2 minutes to run, but this is a highly irregular highrise; and they start digging the hole in April- so time is of the essence.


Thanks in advance for any ideas-
 
if the program does not support multiple CPU's than the only thing you can do is run 4 models at the same time. thus giving you 100% cpu usage.
 
I'm not familiar with ETABS
is the program even multithreaded?
If not, then I would suggest running 4 instances, with staggered 2 hours starts. After the intial ramp up time, you'd be churning out a completed analysis ever 2 hours :cool:
 
Vertigo Acid said:
I'm not familiar with ETABS
is the program even multithreaded?
If not, then I would suggest running 4 instances, with staggered 2 hours starts. After the intial ramp up time, you'd be churning out a completed analysis ever 2 hours :cool:


Excellent idea! Other than I don't know the required changes until the model has finished running.. :eek:
 
jen4950 said:
Excellent idea! Other than I don't know the required changes until the model has finished running.. :eek:
Damn, well there goes that idea :(
Google didn't turn much up about if ETABS has SMP support or not
 
Vertigo Acid said:
Damn, well there goes that idea :(
Google didn't turn much up about if ETABS has SMP support or not
You can run multiple instances of the problem whether it has SMP support or not.
 
mikeblas said:
You can run multiple instances of the problem whether it has SMP support or not.


He understands that, he stated his data sets are dependent on previous runs though, so it's really a non-factor being able to run multiple instances at once.


To the original poster, you're best bet is to call up tech support and see if there is a version of the software that supports SMP.
 
And if there ends up not being SMP support, whine and bitch until you get the fastest single-core you can afford!
 
defakto said:
He understands that, he stated his data sets are dependent on previous runs though, so it's really a non-factor being able to run multiple instances at once.
Oh! Thanks for the correction; I didn't notice the intra-run dependency.
 
Vertigo Acid said:
And if there ends up not being SMP support, whine and bitch until you get the fastest single-core you can afford!

Ha!

I whined and bitched until I got this machine :D I was trying to work on a PIII- it was like trying to cut warm butter with a feather-

So they got me this badass machine- and well, don't get me wrong it's a huge improvement, and I can run 4 instances of ETABS at the same time, and doing other stuff while running doesn't affect it at all; but it would be nice to put all that power to good use. It's like it has ADD- can only focus part of it's effort at a time on a single problem :)
 
Going to the ETABS website, other than mentioning the supported operating systems, there is no mention of recommended hardware that I could find, which seems strange. It looks to be a very advanced structual analysis program, or finite element anlysis.

I would think that any finite element analysis program would require, or utilize the very maximum in available hardware, I know ANSYS does, which supports up to 16 processors, as well as 64 bit Windows.

I know running ANSYS on a general purpose computer w/a single processor w/just just 512 meg of memory has taken me hours. Then if my model fails it's analysis, having to make design changes, then rerun my analysis again, can be frustrating.
 
Because www.2cpu.com is a site dedicated to Xeons, Opterons, Dual CPU type machimes. There are numerous article on SMP and tweaking. If anyone would know about Tweaking Xeons and getting the most out of them, 2cpu would have it.

He said: "Thanks in advance for any ideas" The answer to has question can probably be found there- which is a good idea!!!
 
Well, at least that's a marginally helpful answer. I could just as well have answered "www.google.com" when I saw this thread - after all, there must be something about this on there! The search at 2cpu doesn't turn anything up, and neither does the 2cpu forum search. The 2cpu guys will tell you it's not magic - they can't take a non-multithreaded application and poof it into one.

OP, it might be time to see about a different machine or different software. I know it's not what you want to hear, but with those kind of time constraints it may well be worth it to your company to get it done faster. A fast Athlon 64 or even an FX isn't all *that* expensive, especially compared to a dual Xeon box, and a single fast CPU will be the best option for this. As for other software, I don't know; you may be able to find a common format to export to and import into a demo of other software to test its multithreading ability. You won't be able to reduce run time from 8 hours to 2 - there are only 2 real CPUs there, so maybe 3:15 or so is a more reasonable goal. Hyperthreading only helps so much.

 
dunno if you thought of this Jenn, but contact the software company and get them to offer alternative solutions... works most times when I do that, though I use a finite element grid oriented model..... time of processing = grid size (ie smaller grid size larger processing time)

just a thought

jen4950 said:
Ha!

I whined and bitched until I got this machine :D I was trying to work on a PIII- it was like trying to cut warm butter with a feather-

So they got me this badass machine- and well, don't get me wrong it's a huge improvement, and I can run 4 instances of ETABS at the same time, and doing other stuff while running doesn't affect it at all; but it would be nice to put all that power to good use. It's like it has ADD- can only focus part of it's effort at a time on a single problem :)
 
unhappy_mage said:
OP, it might be time to see about a different machine or different software. I know it's not what you want to hear, but with those kind of time constraints it may well be worth it to your company to get it done faster. A fast Athlon 64 or even an FX isn't all *that* expensive, especially compared to a dual Xeon box, and a single fast CPU will be the best option for this. As for other software, I don't know; you may be able to find a common format to export to and import into a demo of other software to test its multithreading ability. You won't be able to reduce run time from 8 hours to 2 - there are only 2 real CPUs there, so maybe 3:15 or so is a more reasonable goal. Hyperthreading only helps so much.



To go along with what u_m said here.......
Have you tried turning off HyperThreading?

Some applications absolutly hate HT - and they are mostly single threaded apps.

Like yours seems to be.

All HT really does is allow for more efficient multi-tasking. Like switching between applications. If you are running a single threaded application and not trying to do your email, web browsing, taxes, and pr0n0 while it's running then HT doesn't really help you. And with your machine being a dual proc system...... it's kinda overkill having it enabled anyway.

luck

 
Back
Top