| 6:14:01 PM PST - Tue, Nov 6th 2012 |
|
If disk is used it is because the data does not fit in memory anymore, so your example does not apply. NWChem will use as much of the RAM as is available, then resorts to storing the rest on disk, and if there is not enough disk it will recompute. For many systems, disks are very slow so storing in memory followed by recomputing is often faster.
But, do not assume you will have spare RAM to put a disk file system on.
Are you trying to decide what to buy to run NWChem on?
Bert
Quote:Dhaminah Nov 6th 10:24 pmThanks Bert,
Let's neglect the network for the moment and assume we're running the application on a single machine with local storage - disk - across the many cores available. Would the type of storage play a significant role on the application performance. As an example, if we are using the RAM (by creating filesystems from the RAM storage available) vs using regular HDD; is this going to make any difference?
Quote:Bert Nov 6th 2:17 pmThat's a difficult question to answer. Many of our default algorithms use disk and can become I/O intensive. However, those same algorithms can be directed to do "direct" calculations, avoiding disk and becoming compute intensive. A third variable is when you run in parallel, where the network bandwidth and latency can become an issue for certain algorithms.
Thanks,
Bert
Quote:Dhaminah Nov 6th 9:58 pmHi all,
This is my first post on this community. I have a "technical" question about the NWChem software. Is it considered as an I/O intensive application or it is more of compute (i.e. CPU) intensive?
Thanks... |