[Zope] More Pystone benchmarks...

[Michel Pelletier]

> -----Original Message-----
> From: Tony McDonald [mailto:tony.mcdonald@ncl.ac.uk]
> Sent: Monday, October 18, 1999 8:05 AM
> To: Zope List
> Subject: [Zope] More Pystone benchmarks...
> 
> 
> Hi all,
> 
> Q: How comes the Sun boxes are just so bad at Python? is the Sparc 
> architecture *that* bad?

One of the reasons you are seeing much better Apache performance on your
four-by than Zope is probably due to the Python global interpreter lock.
Where multiple apache processes can run on all four processors with no
interlocking, multiple zope threads running in the same process space
all lock around the central python interpreter lock.  While some IO
operations and other things are done unlocked (like Zope DA
communication, if the underying C library was written to release and
acquire the lock) for the most part Zope can't get much advantage from a
multi-processor system.

This is one of the reasons behind ZEO.  ZEO basicly breaks the thread
model out into the multi-processor model, where the processes can be
running locally or on seperate machines.  On your four-by, for example,
you could run 3 or 4 ZEO client processes and one ZEO server process.
When your load fills up, you can simply plug more machines running ZEO
clients into the ZEO server.  No one has tested this scalablility yet
because no-one has come to us with these kinds of load requirements.  It
will happen, in fact, some of our recent customers are gearing up for
this kind of load.

-Michel