Don,

>> I see. That can possibly mean that there's no Win platform performance 
>> regression as I thought, just that the computer was slow.
>>
>> I'll check on my box. Just to be sure: It was run on a single box and 
>> 10.201.200.72 resolved into TCP loopback interface, right?
> 
> Yes, it did.  Thanks for the feedback!

Ok. I've done basic latency test on Win box via TCP loopback (1000 
1000-byte messages as in your test).

For 0MQ/2.0.7 the latency was ~85us, for 0MQ/2.0.6 it was ~65us.

This regression results from removing some kernel-bypass functionality 
(namely lock-free polling) in exchange for more functionality (namely 
allowing for more than 63 threads to use 0MQ sockets).

Severity of the regression depends on efficiency of underlying kernel. 
On Linux it's almost negligible. On Windows... well, the best solution 
would be to optimise the kernel code, but once again... it's Windows :(

Anyway, this was just the standard latency test supplied with 0MQ. Later 
on I'll try to run your test to see whether it won't show some other 
regression.

Martin
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to