[rabbitmq-discuss] High(er) latency with 1.5.1?

[Matthias Radestock]

Holger,

Holger Hoffstätte wrote:
> I was re-running some of the latency tests with 1.5.1 again and noticed
> that the latency is much higher than before, even with TCP_NODELAY.
> 
> Everything on Windows, running MulticastMain with -r 10 -s 100 -i 3 -h
> localhost [...]
> 
> Linux is the same as far as I can tell - almost a factor of 3 higher than
> before (~250-300us after we fixed Nagle). Could this be a consequence of
> the internal scalability changes done for 1.5.0? Or maybe just a
> regression in the test - I think I remember seeing some changes to the
> timing measurements in the changelogs.

I don't think the time measurement code has changed.

The scalability improvement in 1.5.x affect latency somewhat, but not by 
a factor of 3. And I cannot think of any other changes off the top of my 
head that would cause this.

Did you run the tests back-to-back with 1.4.x?


Matthias.