On Tuesday 26 June 2012 19:02:36 Eric Dumazet wrote:

On Tue, 2012-06-26 at 07:34 +0200, Hans Schillstrom wrote:

quoted

This patch didn't give much in gain actually.

With a 100Mbps link it does.

 
I was testing with a patched igb driver with TCP SYN irq:s on one core only,
there was some fault in the prev. setup (RPS was also involved) because now it gives a boost of ~15%

With a 1Gbps link we are cpu bounded for sure.

True.

quoted

The big cycle consumer during a syn attack is SHA sum right now, 
so from that perspective it's better to add aes crypto (by using AES-NI) 
to the syn cookies instead of SHA sum. Even if only newer x86_64 can use it.

My dev machine is able to process ~280.000 SYN (and synack) per second
(tg3, mono queue), and sha_transform() takes ~10 % of the time according
to perf.

My test machine is not that fast :-(
I have only 170.000 syn/synack per sec. and sha_transform() takes ~9.6%
have seen peeks of 16% (during 10 sec samples)

With David patch using jhash instead of SHA, I reach ~315.000 SYN per
second.

I have similar results from ~170k to ~199k synack/sec.

BTW, 
cookie_hash() did not show up in the perf results, (< 0.08%)