Hello Gerald -
The "used" memory reported by /proc/meminfo is apparently not
a meaningful measurement of non-freeable memory when there are
large values in /proc/slabinfo. There are some XFS and page_buf
entries in there which can get quite high. However they go right
back down if something else really needs the memory. So I don't
think there's a leak. I'd sure like to understand what's really
going on a bit better, but the only thing which seems broken is
our ability to easily get an idea of how much free-able memory
is left :-(
Be seeing you,
- Sidik
<Steve Lord wrote:>
|
|> I ran the 1.0 redhatish kernel for months with no major problems other than
|> a few data corruption bugs on dirty recovery. I was hoping that upgrading to
|> 1.0.1 would fix those issues, but now I seem to be running out of memory in
|> the space of about 2 days. The "used" ammount of memory seems to be higher
|> than the sum of all processes, so it seems like something is leaking userspac
|> e
|> memory. Anyone else having these problems? Any fix available?
|>
|> -- Gerald
|
|Hmm, maybe there is an interaction with changes from redhat, is the machine
|actually behaving worse? The oss website has been up and running on a 1.0.1
|XFS kernel for nearly 31 days now and that machine is pretty busy. However,
|the kernel happens to be from just before the release and is still 2.4.2
|based.
|
|Can you post some more info on what it is you are seeing.
|
|Thanks
|
|
| Steve
|
|
|