Not sure if this is too trivial for this group… but as a performance person on 
AIX I generally look to see how often is the sweeping hand coming by and how 
much RAM is it actually collecting…

On Linux you want to watch with something like this:
watch -n1 'grep -E "pgfree|pgsteal|pgscan|nr_free_pages" /proc/vmstat'

To find out how much pressure your server is under you want to look to see how 
many page scans are being performed (is counter incrementing quickly)… that 
means the system WANTS memory, versus how large are the  page steals how much 
memory did it get…

Typically, you’d divide the steals by the scans to come up with a pressure 
ratio.

If your server is using ALL of its RAM say for cache hits etc, that can be 
efficient, so long as the ratio is low to how much RAM it demands and how often 
is it trying to get free memory.  If memory is just lazily being occupied the 
inactive page count will be high, and as soon as a scan demands memory it will 
free a TON of memory.  But if the memory is marked active, and the scan demands 
memory it won’t be able to free memory, so you’ll end up with many scans and 
much smaller freed memory.

Hope that helps.


From: gpfsug-discuss <[email protected]> On Behalf Of Jonathan 
Buzzard
Sent: Monday, February 16, 2026 8:35 AM
To: [email protected]
Subject: Re: [gpfsug-discuss] swapped_warn event




On 16/02/2026 11:43, Jan-Frode Myklebust wrote:

>

> It seems like someone thinks that linux servers should never use any

> swap. This swapped_warn triggers if you've used more than 50 MB of

> swap space.  I find this silly..



That's a polite way of putting it.



Anecdotally, if memory starts to get tight Linux will preventatively

push pages to swap "just in case", and might never actually use it.

Consequently having a warning if more than 50MB of swap space is used is

as useful as a chocolate teapot.



If you are going to warn about swap being used, then it needs to be

because the kernel was actually shuffling memory between disk and RAM

not because it pushed some pages to disk just in case.



On an HPC system where the compute nodes are frequently near maximum RAM

usage all I unsurprisingly have a load of spurious warnings.



>

> You can tune it using something like "mmchconfig

> mmhealth-gpfs-swap_alert_threshold_kb=2000000 --force", but I wouldn't

> want to pollute my config with such settings.. Maybe just disable it

> using "mmhealth event hide swapped_warn".

>



I am debating having a script automatically run "swapoff -a ; swapon -a"

on the nodes if these warnings are seen :-)





JAB.



--

Jonathan A. Buzzard                         Tel: +44141-5483420

HPC System Administrator, ARCHIE-WeSt.

University of Strathclyde, John Anderson Building, Glasgow. G4 0NG



_______________________________________________

gpfsug-discuss mailing list

gpfsug-discuss at gpfsug.org

https://urldefense.com/v3/__http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org__;!!F9svGWnIaVPGSwU!o15h3llP_t-xAwde_PAYRCF1VYCK7qhlPXsK1TCZNCjGQSiX71ae4hsr_lQQzfzDbWX_iltIVIDP05jTcSf95e2pUMfriO6pIymIMw$<https://urldefense.com/v3/__http:/gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org__;!!F9svGWnIaVPGSwU!o15h3llP_t-xAwde_PAYRCF1VYCK7qhlPXsK1TCZNCjGQSiX71ae4hsr_lQQzfzDbWX_iltIVIDP05jTcSf95e2pUMfriO6pIymIMw$>
_______________________________________________
gpfsug-discuss mailing list
gpfsug-discuss at gpfsug.org
http://gpfsug.org/mailman/listinfo/gpfsug-discuss_gpfsug.org

Reply via email to