On Tuesday 07 August 2001 03:00 pm, Matt Stevens wrote:
> I have a box that I recently upgraded to Squid 2.4 Stable 1. I also
> upgraded the linux kernel to version 2.4. The box previously used little
> or no swap (~3M) whereas now it's using over 100M of swap.
>
> Is this normal for the 2.4 kernel or is this something to do with Squid?
>
> The box is an Athlon 1Ghz, 1.5G RAM, 500M swap partition. One 30G system
> disk, with two 45G cache disks.
This is normal behavior as of kernel 2.4.5. Blocks are not removed from
swap after getting swapped back in to avoid fragmenting the swap
partition. This can, and does, present a problem on systems where swap <
2xRAM.
Possible fixes:
* Repartition so swap >= 2xRAM. The maximum swap partition size is 2GB,
so you will need more than one -- maybe 1.5 GB on each disk.
* Disable swap. The kerndev folks say this is stable (mostly for embedded
systems). I'm skeptical.
* Try 2.4.0-test12. It doesn't like the rapid spawn/kill life of our
other systems, but it has been very solid on our squids.
As I type this, our main web server (kern 2.4.7) just died -- the third
time in a month -- so my opnion of the later 2.4s is extremely low right
now.
-- Brian
Received on Tue Aug 07 2001 - 13:30:51 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:01:30 MST