* [RFC][0/2] remove atm_find_ci(); convert vcc_sklist to hash
@ 2003-10-08 12:46 chas williams
2003-10-08 13:58 ` David S. Miller
0 siblings, 1 reply; 2+ messages in thread
From: chas williams @ 2003-10-08 12:46 UTC (permalink / raw)
To: netdev
the following two patches (which should be applied to a current 2.6 tree)
obseletes atm_find_ci() which races the vpi/vci allocation (since the
actual allocation is delayed until after the entering the drivers open()
routine). all drivers now use the built-in find_ci(). to address
performance issues the second patch converts the vcc_sklist to a hash
based on the lower 5 bits in the vci. should this be tunable via config?
for most users (< 100 vccs) this should be fairly efficient. on a side
note, signalling/ilmi/control vccs are all in bucket 0. the control
vccs probably should be given a non-zero vci to make this 'better'.
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [RFC][0/2] remove atm_find_ci(); convert vcc_sklist to hash
2003-10-08 12:46 [RFC][0/2] remove atm_find_ci(); convert vcc_sklist to hash chas williams
@ 2003-10-08 13:58 ` David S. Miller
0 siblings, 0 replies; 2+ messages in thread
From: David S. Miller @ 2003-10-08 13:58 UTC (permalink / raw)
To: chas3; +Cc: chas, netdev
On Wed, 08 Oct 2003 08:46:26 -0400
chas williams <chas@cmf.nrl.navy.mil> wrote:
> the following two patches (which should be applied to a current 2.6 tree)
> obseletes atm_find_ci() which races the vpi/vci allocation (since the
> actual allocation is delayed until after the entering the drivers open()
> routine). all drivers now use the built-in find_ci(). to address
> performance issues the second patch converts the vcc_sklist to a hash
> based on the lower 5 bits in the vci.
These changes look fine, I'll add them to my tree.
> should this be tunable via config?
I don't think so, at least not now. My advice is, keep it a constant
for now and if people run into perf problems and the hash table size
is determined to be the cause you can investigate how to choose
different hash table sizes.
This is exactly what we did with the TCP hashes.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2003-10-08 13:58 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-10-08 12:46 [RFC][0/2] remove atm_find_ci(); convert vcc_sklist to hash chas williams
2003-10-08 13:58 ` David S. Miller
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).