Re: excessive swap-in time

phil-news-nospam@xxxxxxxx wrote:

I sometimes encounter swap-in times for some heavy processes (usually the
Firefox browser) that can take a few MINUTES to complete. Reading all of
the swap space sequentially only takes 40 seconds. I think what really
is happening is that the process is swapping back out as it needs space
to swap in, even though the total VM size is less than half of real RAM.

In my experience with recent Linux, swap performance gets terrible after the
system has been up for awhile. The first heavy swap out goes along at
blistering speeds, but after a few months both swap in and swap out starts
behaving the way you describe.

It would be nice if there was a "swap trace" feature whereby I could see
the swapping actions taking place. Of course it could be a problem as the
process reading such a trace would affect things.

Maybe the block_dump facility could be of help?

echo 1 > /proc/sys/vm/block_dump

Please shut down syslog/klogd/minilogd first, you don't want to be logging
syslog's logging of the logging of the logging of the logging of the....

I'd like to know how
often the same page has to get swapped in over and over.

I suspect that the on-disk layout got worse at some point, needing much more
seeking than before.
.

Relevant Pages

  • Re: RFT: updatedb "morning after" problem [was: Re: -mm merge plans for 2.6.23]
    ... that user file system is on the same drive as the swap partition ... with the swap partition accesses might overwhelm all other performance ... moment (if I'm reading the code correctly) the system has to have close to ... workable - its small and does a task that is best done in userspace. ...
    (Linux-Kernel)
  • Re: RFT: updatedb "morning after" problem [was: Re: -mm merge plans for 2.6.23]
    ... If the problem is reading stuff back in from swap at the *same time* ... that user file system is on the same drive as the swap partition ... Perhaps we could have a 'wake-up' command, ...
    (Linux-Kernel)
  • Re: multimeter question
    ... > When i measure resistances with my dmm, i get a different reading if i ... > swap the leads around. ...
    (sci.electronics.repair)
  • Re: 4 Gig are missing
    ... Your swap is evidently about 2gb, so accounted for partitions equal 187GB. ... Errm...try reading the data supplied, especially for hda3: ... You're right - I was bombarded with too much extraneous information. ...
    (alt.linux)
  • Re: Out of disk space problems when using 32M /boot
    ... I have taken your advice, ... I did a bit of reading and found out that the 1st partition always ... failed when my seperate partition was only 32M. ... I had a read about tmp on swap, but I'm still not sure how to go about ...
    (comp.unix.bsd.freebsd.misc)