Linux (virtual) memory tuning

[Jerome Henry]

Hi everyone,

I'm working on linx memory tuning.

Has any of you tuned kswapd (tries base, tries min and swap cluster) or bdflushd ?

Could you provide any comparative benchmark on any of them ? I'm preparing a lesson for my students on that, and only have my poor notebook as a benchmark reference, so would be glad to get some more stuff, or even pointers on the net, to show what are performances differences on moving values for them...

Tks !

J

[dirk dierickx]

Old but still nice: http://www.tldp.org/LDP/solrhe/Securing-Optimizing-Linux-RH-Edition-v1.3/chap6sec68.html

[Jerome Henry]

Thanks for the link ! I had this url.

Mmmm... not much interest on that point everyone, huh ?
Let's put things differently :
ON followinf the former link, you'll get informations on bdflush, which determines when pages have to be consiederd as being to be written to swap and kswapd, which determines how and when pages to swap will be transfered to disk.
Following another link, which gives some tests done on those (among others) for Oracle optimization, you'll see how to set up different values for them.
http://www.puschitz.com/TuningLinuxForOracle.shtml#SizingSwapSpace
If you have a spare machine, I mean not your precious production server but a machine on which you could tune these values, which are not much risky to change if you respect suggested values, I would be glad to give a bunny to anyone who could provide compared values between original standard kernel settings and improved settings based on last pointer for example.

Tks !

J

[Steven E. Protter · Accepted solution]

Jerome,

I have a fedora machine thats in for eventual configuration as a backup dns/emergency web server for my business. the backup site won't be up for 30 days or more so I don't need it right away.

If you still have my email, let me know or just use the customer contact form at http://www.isnamerica.com and zap me here so I know when to look.

I'd be happy to arrange access so you can try your theories. Its going to be a red hat 3.0 es testbed. We'll just have to coordinate when it gets zapped.

SEP

[Jerome Henry]

Thanks a lot Steven,
You are always so helpful. In between, I luckily could alter some configs on some of our university servers, plenty of configs and plenty of tests.
I'll add the benchmarks to my next publication.
Thanks again !

Yours J (of course still keeping your e mail :]] )