Skip to contentSkip to conversation

E1000 - page allocation failure - saga continues :( message 1

DISCUSSION Paul Nowoczynski 1 participant 1 message May 11, 2005

lore
Share & tools
lore
#0 lore raw

I think the problem is related to dma activity by the e1000 and the page cache hoarding all the pages in the system. I've been using min_free_kbytes to get around this - setting it at 131072. Even at that amount I still see the page_alloc errors, raising the limit to 262144
actually lowers the amount of memory the system reserves (to about 90MB)! I've seen this
on both 2.6.7 and recently 2.6.11.

I'd like to see more control over the page cache. My machine do tons of disk IO and really I find it a waste that 1.8 GB of memory is sitting idle in the page cache, meanwhile the kernel is unable to free it fast enough when the ethernet card needs to dma.

If I'm missing something please let me know.
paul

Quoted 11 lines · [email protected] wrote:
Quoted 7 lines · Nick Piggin - [email protected] wrote:
Quoted 2 lines · [email protected] wrote:

It would be nice with a "cleaner" solution though.

What kernel are you using?
Are you doing a lot of block IO as well?

Quoted 6 lines

I am using 2.6.11.8.

Yes, the server is a fileserver for both the internet (~10Mbit) and internally (1Gbit e1000). Hardware is pretty old so is pretty heavily loaded and with 256MB RAM.

OK, well there are some patches in 2.6.12 that should make things slightly better, and then some more patches in -mm (not sure if they'll make it for 2.6.12) that should make things slightly better again.

Basically they work towards reducing the memory allocation "priority" for block IO requests, in relation to networking and other atomic allocation requirements.

If you can't test the latest -mm, or 2.6.12-rc4, then wait for 2.6.12 and 2.6.13 and check back on the problem.

Thanks,
Nick

Signature · 3 lines
SUSE Labs, Novell Inc.

-