[LARTC] ESFQ not so fair?

[Corey Hickey]

Patrick McHardy wrote:
> Andy Furniss wrote:
>> Corey Hickey changed his esfq to use jhash for dst/src/fw - copy of his
>> announce below.
>>
>> Andy.
>>
>> Corey Hickey wrote:
>>> So, I wrote an alternative hash function. It's quite simple, and as long
>>> as the range of input values is smaller than the hash table (default
>> 1024,
>>> up to 16384), collisions will not happen at all. See the updated README
>>> file for more details.
> 
> Using jhash is a probably a good idea, the "improved" hash is broken
> and will cause reordering in some circumstances:
> 
> return (h - q->dyn_min) * (q->hash_divisor - 1) / q->dyn_range;
> 
> dyn_min, dyn_max and dyn_range, as their name suggests, are adjusted
> dynamically, so the hash function changes whenever one of these values
> changes, resulting in reordering of packets belonging to a single flow.

That should stabilize after it's been running a while and has seen the 
normal range of IP addresses. Anyway, I agree, it's not very good.

Working on ESFQ some more has been on my long-term TODO list, but what 
with getting distracted by mplayer development I didn't get around to 
it... and now I have 1.2 real jobs and 1.0 girlfriends and don't have 
time for much programming. :)

If any of you want to send patches to this list and they don't look bad 
to other readers of the list I'll happily apply them and make a new 
release. Other than that, I can't help you much for now.

Thanks,
Corey