Unable to use full RAM
 

Hi all,

Well, I am running a complex simulation on CFX with multiple phase and is running on all 48 cores but I am not sure why its not using more ram and speed it up.
I have 256GB ram , out of which it is only using like 35GB and is running quiet slow.
I m running the simulation on MPCH2 Distributed parallel.
I want to know how we can input, memory allocation which it should use to run just like we do for cores.

Regards,
MAK

Hi all,

I was searching thru internet n I found out that above certain fixed ram memory, windows dosent recognise more memory.
Like WIndows XP has a limit of 4GB, so if u use windows XP with say 8GB, it wont recognise 4GB memory unless specified in boot.ini file.
http://msdn.microsoft.com/en-us/windows/hardware/gg487508.aspx
I reckon this could be the issue too but i am not sure.
Can sum1 help me in this??...
I m using Windws server 2008 R2 with HPC packs

Regards,
MAK

i don't think this is the problem. usually ram doesn't speed up the solution, it just define the size of the problem that can be solved. if you are using 35 gb of ram id say you have something around 10-20 M cells problem give or take 5 M depending on how many species and turbulence model is in place. correct me if I'm wrong.

of course the above is valid if the whole system is not broken and is not swapping (writing on disk) but swapping is just lot-of-years-to-get-results-slow compared to just quite-slow. have you benchmarked your problem and SW to see if it is worth to use so much cores?

yeah i may not b the problem ... i will just try to run 4 different simulation parallely n see if it picks up d ram. from wht i knw, core helps pick up big size simulations and ram helps in running it fast.
I ran 28M elements with 24 cores but it gave me error of memory problem, so i thought of running lesser elements with more core. hence, i m running 12M elementswith 48 cores which is running perfectly fine.
My only worry is to why its not picking up RAM, not even with 28M or 12M elements. The RAM usage looks same to me in both the simulation.

Nah its not tooooooo slow, just not upto d mark.
Actually I just go tthis system, so m not sure both the system configuration n all... I worked ou td ram usage for 28M which was 24 core. but with 12M i just took max cores to see how it reacts :)

Still m not sure whats goin on in d system. Let me run 4 diff simulations n c :)

it's actually the opposite.

then there is something wrong. i assume the system is a single board (no cluster)... the ram usage should be different for the two cases.

i don't use whs, so i can't help. but your test should give some hints anyway.

Thanks for the reply sail.

Well I m out of work for few days.
Will test it once I m back.

Thanks for the suggestions Sail,

I worked it out and now its working on all cores and with greater RAM usage.

just for my curiosity, and for the forum's future memory, what was the problem? what have you done to solve it?

Thanks.

Found out tht ,

local parallel uses 100% CPU usage available whereas Distributed Parallel divides the simulation equally among cores n uses more physical memory to run it...
So, even if the problem is big n you run it on local parallel, it wont take much physical memory