Host38 is being upgraded and rebooted. More details in a few...

Ok. Host38's kernel has been upgraded and Linodes are restarting now.

There's a nasty bug in 2.6.9 which causes UML processes to not release their memory. So, each Linode reboot potentially did not release the memory it was using. This caused the host to kick in the OOM killer, which is why you guys were experiencing problems.

The new host kernel includes the cfq-time-sliced disk sched, so you should see a decent disk performance increase.

Sorry for the trouble.

-Chris

Will this affect other hosts too? Should we expect a reboot?

-- jp

Can we get a report from the Linode users on host38 about how significant an impact this cfq-time-sliced disk sched change has on performance? I'd love to hear that it mitigated the load spike problem, which still occurrs sometimes for me (although I'm not on host28, but if that kernel helps host28, then I'm all for upgrading host31 too).

It was very slow last night right after he fixed it, now it feels much faster then it used to be.

There are no other hosts running 2.6.9, or any of the 2.6.10-rc based patches that were also affected. However, they'll still be a round of host kernel upgrades coming. Announcements and details are forthcoming.

-Chris

Performance seems to have increased a lot, but I have upgraded the ram of the system from 64 to 128, so i don't know how the the two things participated to the overall improvement.

bye