Thomas Jacob wrote:
On Thu, 2008-07-03 at 14:42 +0200, Patrick McHardy wrote:
Please work this out among yourselves. I think we should just pick
the faster one since both don't look very intrusive.
Well, Jesper's version doesn't build any extra data-structures,
so you save time&space there, and the offsets stuff is really
only needed once, so the chain_index rebuilding penalty
doesn't play a role.
And since about 2/3s of time it takes to load my 50k Chains now
is system time, it's probably irrelevant whether
0 + O(n * (log(n/40)+40)) is sometimes larger than O(n+n/1024) +
O(log(n)).
Mine vs. Jesper's:
iptables-restore (50k chains, 120k rules), average for 10 runs:
User: 2.558 s - System: 8.672 s - Total : 11.222 s
vs
User: 2.622 s - System: 8.520 s - Total : 11.140 s
iptables -vnL SOMECHAIN (2 entries, with the above ruleset in kernel),
average for 20 runs
User: .094 s - System: .363 s - Total : .455 s
vs
User: .085 s - System: .389 s - Total : .472 s
Those numbers are all within the standard deviations of each other,
so there is no difference for practical purposes, I think :-)
I would use Jesper's patch.
OK, thanks for the numbers and sorting this out so peacefully :)
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
