[viff-devel] What to benchmark

[Martin Geisler]

ivan at daimi.au.dk writes:

Hi!

> I think that looking at the two bottlenecks as suggested is a very
> good idea and should have the highest priority. We don't know nearly
> enough about how much time Python in general and Twisted in
> particular is stealing from us. In principle, we could be in a
> situation where it is hopeless to get the performance we should be
> able to get unless we ditch Python and implement Twisted ourselves
> in C++.

Yes, we need a much better understanding of what is going on. Only
then can be start talking about what kind of performance it is
realistic to hope for, and then look at how to achieve it.

> Note that I am not saying we are in that situation, in fact I don't
> think so - but I am saying that it is important to find out ASAP!

Agreed! I would be very happy to hear suggestions as to how we can
measure things in VIFF and/or Twisted.

I have tried looking at the idle time in the reactor (Twisted's event
loop) and found that it blocks on the select call only 40 times when
multiplying 4000. It starts by sleeps 20 ms in 23 select calls, and
then alternates between the following two steps:

* 315-317 multiplications finish
* sleeps 10-11 ms in 2 select call

I don't know yet what the number 315 corresponds to...

-- 
Martin Geisler

VIFF (Virtual Ideal Functionality Framework) brings easy and efficient
SMPC (Secure Multi-Party Computation) to Python. See: http://viff.dk/.