Forums » General

Upgraded Server Kernel

Jul 19, 2005 Phaserlight link
See the news for details.

Just how is that new kernel on majikthise working out? Any noticable effects on server performance and/or laaaaaaag?
Jul 19, 2005 incarnate link
I would not expect any improvements in that. I doubt there were even any IP-stack or significant architecture fixes between 2.6.5 and 2.6.12 that would impact performance. This is a stability upgrade only.

Having mysterious stray lisp processes that we couldn't kill on 2.6.5 was scary. Scary is bad. Baaad.
Jul 19, 2005 igrok link
If you are interested in seeing lag reduction, you could consider the Con Kolivas patchset which includes a different scheduler. I can't guarantee that it would improve network performance, but I have seen significant response improvements on systems under heavy load. At work we have seen some *significant* response issues on the newer kernels (2.6.10+). If a system is really loaded, simply logging in on a console can be painfully slow.

You could always try it out on the test server :)
Jul 19, 2005 roguelazer link
cfq I/O scheduler works well for me.
Jul 19, 2005 greengeek link
I've played around with the various scheduler patches a bit on my own machine, though it's a bit of a dark art to figure out which ones produce better throughput, and which ones "feel" faster. Plus there are two different sets, the process scheduler and the i/o scheduler. Not that it's a bad thing to play with, but I don't think anyone wants the devs rebooting majikthise 10 or 15 times to compare kernel patches ;)

Changing the process scheduler can either make a machine faster at pushing a single major process (great for a single purpose server), or more responsive to multiple competing processes (Better for a desktop or multi-role server).

Unless the server is task-swapping very often (doubtful) or running into funny I/O bottlenecks between memory, CPU and storage, I doubt that patching the kernel would produce a noticable difference from our perspective, and is likely to introduce more instability.
Jul 21, 2005 a1k0n link
Yeah. Changing the IO scheduler might have a miniscule impact on performance, but changing the userland code I wrote would affect things much more heavily unless the IO scheduler is horribly broken.
Jul 22, 2005 LeberMac link
Perhaps post these happenings on the "In-Progress" section?
Tuesday, March 22, 2005 What's Up In The Wild World of VO? Hi, there's been a big lack of In Progress updates...

Or maybe just wait until the hive update goes live and then write another big article...

Perhaps now that I can't actually PLAY the game I can partially assume Genka's role and just post pithy comments.
(Watch them lock me out now....)