Re: how to know when much page reclamation

[Ed L Cashin <ecashin@xxxxxxx>]

Eugene Teo <eugene.teo@xxxxxxxxxxxxx> writes:

> <quote sender="Ed L Cashin">
>> Hi.  I can choose one of two different related things in userland, and
>> one takes more physical memory than the other but runs faster.
>
> you mean one memory intensive, one io intensive?

Nope.  :)  

Strategy "A" uses more page tables (pinned in RAM) and runs fast,
while strategy "B" uses fewer page tables and runs somewhat slower.

>> Looking at the number of free pages in the system doesn't really tell
>> me whether memory is "tight" or not.  There could be plenty of pages
>> being used for caches, and at the first sign of memory pressure, those
>> caches may shrink.
>> 
>> The best way I can think of to know whether memory is really tight is
>> to measure how often shrink_zone is getting called.  (I'm looking at a
>> 2.6.0-test11 kernel.)
>> 
>> I can add some light instrumentation to track the rate at which
>> shrink_zone gets called and a system call to make that rate visible to
>> userland.  Is there an existing way already in place or a better way
>> to know when memory is tight?
>
> cat /proc/vmstat, see pgscan. but the stat is commulative. 

At a glance, that looks like just the thing, with stuff like "pgsteal"
and "pgscan".  Thanks!  I will look into it.

I'm interested in system-wide memory pressure, so that's OK.  I can
sample periodically and observe the rates over time.

-- 
--Ed L Cashin            |   PGP public key:
  ecashin@xxxxxxx        |   http://noserose.net/e/pgp/


--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/