|
[Sponsors] |
Shared memory Processor WS vs Distributed computer system (clusters) |
|
LinkBack | Thread Tools | Search this Thread | Display Modes |
January 24, 2010, 17:00 |
Shared memory Processor WS vs Distributed computer system (clusters)
|
#1 |
New Member
Frank
Join Date: Jan 2010
Posts: 2
Rep Power: 0 |
Hi, I am planning to invest some money to do my research using Fluent and Flow3D software. Size of my applications could be varied from 10-50 million cells and sometimes I may need to run 3 or more simulations in parallel. When I inquired about above two options, I got mixed opinion from venders and different CFD users. Please help me to finalize my decision with your opinion. There is no limit for budget.
I may also need to develop many models at the same time with two or more attached workstations. If possible, also please suggest the brand and model of processors, graphic cards, etc. |
|
January 25, 2010, 05:57 |
|
#2 |
Senior Member
Join Date: Dec 2009
Posts: 131
Rep Power: 19 |
If this is just for your research (university, master or phd??) and you will not be running this for hundreds/thousands of jobs per month or you will be the only primary user, the Cray CX1 may be just what you are looking for. It has compute, storage, and visualization blades, all in one nice box (nicely balanced system). If you need some more compute power, you can add another CX1 to it!
http://www.cray.com/Products/CX/Systems.aspx There is also the Appro - Ready-To-Go Cluster Series http://www.appro.com/product/ready-to-go_clusters.asp |
|
January 25, 2010, 06:37 |
|
#3 |
Senior Member
Join Date: Dec 2009
Posts: 131
Rep Power: 19 |
one other note.
you did not mention if these will be steady/unsteady, the turn around time for calculations, or how many licenses you will have access to. This would also play a part in how to setup these jobs and the hardware requirements to meet these needs. |
|
January 25, 2010, 10:23 |
|
#4 |
New Member
Frank
Join Date: Jan 2010
Posts: 2
Rep Power: 0 |
mjgraf, thank you for your reply. To answer your questions; I am in a university setup. Almost all the simulations will be at transient state. I will at least have 2 licenses for each software (I think in order to setup 2 problems simultaneously, you need 2 licenses). Turn around time for models with 5 million cells that I am running now takes about 5-6 hrs. I will visit above website and check it however, if you have more info please post! thanks again. Sekula
|
|
January 25, 2010, 15:27 |
|
#5 |
Senior Member
Join Date: Dec 2009
Posts: 131
Rep Power: 19 |
how many cpus are you currently running for these 5 million cell jobs?
5 to 6 hours for a 5 million cell transient seems too good to be true unless you are throwing a ton of cpus at the job. of course i do not know the details of the job. The transient jobs we are running take days to compute. |
|
Tags |
clusters, smp |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
CFX11 + Fortran compiler ? | Mohan | CFX | 20 | March 30, 2011 19:56 |
DecomposePar unequal number of shared faces | maka | OpenFOAM Pre-Processing | 6 | August 12, 2010 10:01 |
Which is faster, distributed or shared data? | Cho | Siemens | 2 | June 17, 2004 17:36 |
Run a shared memory parallel version on a network? | Cheng MuLin | FLUENT | 1 | October 23, 2001 04:58 |
ECC Memory Needed for Clusters? | Jonas Larsson | Main CFD Forum | 1 | January 17, 2001 09:08 |