On Fri, May 29, 2015 at 09:45:38AM -0400, Donald Sharp wrote:
> To Continue with Paul's point.  I believe that there are allot of
> performance gains to be had by running perf and looking at the output to
> see where our algorithmic/data structure choices are wrong and fixing those
> first before diving in and rewriting for multi-threaded code.

I don't think anyone had suggested multithreading yet.

Some people would prefer one process with multiple VRFs to save resources.

Some people would prefer multiple processes with one VRF each.

Certainly one process with threads might be able to match the scalability
of the multiprocess model while working efficiently on low end hardware
too, but multithreading can also be much harder to get right (with
multiple single threaded processes you know exactly what is going on in
the process).  Certainly multiple processes give nice isolation between
things.

It certainly seems that having single threaded processes that can do
multiple VRFs but can also be run as multiple processes would make
everyone happy, although it does mean there are two ways of running
things that would have to be supported in the code.

-- 
Len Sorensen

_______________________________________________
Quagga-dev mailing list
[email protected]
https://lists.quagga.net/mailman/listinfo/quagga-dev

Reply via email to