Re: [HPDD-discuss] OSC and buffer cache throttling
On Feb 3, 2015, at 3:41 PM, Drokin, Oleg
<oleg.drokin(a)intel.com> wrote:
I tracked the change to http://review.whamcloud.com/2514 which was landed in Oct 2012,
but in it's original form if you had less than certain amount of RAM, then it
defaulted to 3/4.
Is your 2.5.3 client have relatively little RAM by any chance?
Here are the stats from a few different client systems.
Client #1 (Linux cluster login node):
- Lustre 2.5.3
- 64 GB RAM
- max_cached_mb = 48 GB
Client #2 (External Cray login node):
- Lustre 2.5.1
- 256 GB RAM
- max_cached_mb = 128 GB
Client #3 (Internal Cray login node):
- Lustre 2.5.1
- 64 GB RAM
- max_cached_mb = 32 GB
Notice that client #1 and #3 have the same amount of RAM but different defaults for
max_cached_mb. (Not that it’s a problem. I just never noticed the difference before I
took a closer look today.)
--
Rick Mohr
Senior HPC System Administrator
National Institute for Computational Sciences
http://www.nics.tennessee.edu