Re: HalfSipHash Acceptable Usage

[Eric Dumazet]

On Tue, 2016-12-20 at 22:28 -0500, George Spelvin wrote:
> > I do not see why SipHash, if faster than MD5 and more secure, would be a
> > problem.
> 
> Because on 32-bit x86, it's slower.
> 
> Cycles per byte on 1024 bytes of data:
> 			Pentium	Core 2	Ivy
> 			4	Duo	Bridge
> SipHash-2-4		38.9	 8.3	 5.8
> HalfSipHash-2-4		12.7	 4.5	 3.2
> MD5			 8.3	 5.7	 4.7

So definitely not faster.

38 cycles per byte is a problem, considering IPV6 is ramping up.

But TCP session establishment on P4 is probably not a big deal.
Nobody would expect a P4 to handle gazillions of TCP flows (using a
32bit kernel)

What about SHA performance (syncookies) on P4 ?

Synfloods are probably the only case we might take care of for 2000-era
cpus.