mysql-proxy-discuss team mailing list archive

[Kay Röpke <Kay.Roepke@xxxxxxx>]

Hi!

On Sep 7, 2009, at 7:49 PM, Adrienne Porter Felt wrote:

> We seem to have fixed the problem.  When we use a Python profiler that
> computes CPU time as opposed to wall time, we get the expected  
> results (a
> consistent slowdown for the proxy which decreases as the result set  
> size
> and/or complexity of the query increases).

I'm just a bit surprised that measuring wall clock consistently shows  
results going in the other direction,
actually. The only explanation I can come up with is this running all  
on the same machines where the active
threads exceeds the CPU/core count of said machine. Was this the case?  
Otherwise wall clock time should be
higher, but exhibit the same general trend, no?
But good that you could find and solve the problem!

> The only remaining oddness is that increasing the number of  
> connections
> going through funnel still shows a speedup (that is, processing 1
> query when 250 connections are open to the proxy and 1 connection is  
> open to
> the backend is *faster* than processing 1 query when 10 connections  
> are open
> to the proxy).


Maybe Nick has an explanation for that, I can't see an obvious reason  
why this should be true.
It's worth writing a benchmark for that, though, and try to see if we  
can come up with profiling data that
shows it (and then hopefully lets us understand it).

Thanks for the valuable data!

cheers,
-k
--
Kay Roepke
Software Engineer, MySQL Enterprise Tools

Sun Microsystems GmbH    Sonnenallee 1, DE-85551 Kirchheim-Heimstetten
Geschaeftsfuehrer:    Thomas Schroeder,  Wolfang Engels,  Wolf Frenkel
Vorsitz d. Aufs.rat.: Martin Haering                    HRB MUC 161028