Re: 2.6.33 problems

[Eric Dumazet]

Le mardi 02 mars 2010 Ã 14:42 -0800, Stephen Hemminger a Ãcrit :


> The SNMP MIB stuff does allocate a disproportionately large amount of percpu
> data. Looks like 2 tables per MIB.

But did this percpu usage grew with 2.6.33 ?

In my understanding, ipv6 always wanted a lot of percpu data for its
MIBS (including per device MIBS), the big increase occurred with commit
96793b482540f3a26e2188eaf75cb56b7829d3e3
from David L Stevens, when RFC 4293 was implemented.
On IPV6, each net device needed 4096/8192 additional bytes of memory per
possible cpu.

An attempt to reduce SNMP mibs by 50% was attempted one ago, but not
completed. I'll try to find some time to resurrect the patch.

[RFC] percpu: convert SNMP mibs to new infra

(use one single table, for both user/bh contexts on x86)

http://lkml.org/lkml/2009/4/1/401


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/