I am working with diskless servers and monitoring 'free' output. I have observerd several application crashes when the cache memory size becomes equal to the available memory size. I know cache is normally used for holding data for disk i/o - and that is all the literature names. In a diskless system, what else is it used for?
The system does run NFS but (supposedly) that is not in constant use. I need to determine how much memory has to be held in reserve and not allocated directly to the applications - they hog as much as they can get. There is no swap (no drives). Apparently, in this arrangement, forced memory recovery from cache is detrimental to the applications. Buffer memory is zero.

Set your vfs_cache_pressure (/proc/sys/vm/vfs_cache_pressure) high so that RAM caching doesn't happen. Also change 'swappiness' value, if you have more RAM.

Have a script to set '3' to /proc/sys/vm/drop_cache, so that if any cache will be cleared giving room for applications.

1 members found this post helpful.

Thanks hi2arun.

I am still interested in what could be using the avilable memory though. Any way to tell?