produce memory overhead in bazaar

[John Arbash Meinel]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 11/29/2010 3:00 AM, Martin Pool wrote:
> On 26 November 2010 21:44, Lukas Diekmann
> <Lukas.Diekmann at uni-duesseldorf.de> wrote:
>> Hi there,
>>
>> my name is Lukas and i am working on optimizations for pypy. Pypy is a
>> python interpreter written in python itself and aims to simplify,
>> improve and speed-up python interpretation. For my master's thesis I am
>> writing several benchmarks based on real programs that use a lot of
>> memory. Thus I'm hoping to find out which datatypes produce the most
>> overhead so I can extend the pypy interpreter and jit by implementing
>> additional datatypes that are optimized for their purpose (for example a
>> list containing integers only).
>>
>> I think that your program could be a candidate for benchmarking because
>> it seems to eat a lot of memory. But since I am new to bazaar it is hard
>> for me to tell and produce a situation where it uses the most memory. So
>> I wanted to ask you, if you could tell me a scenario that is worth
>> looking at and how I can reproduce it.
>>
>> If you want to know more about this project have a look at
>> http://morepypy.blogspot.com/2010/08/call-for-benchmarks.html
>> <http://morepypy.blogspot.com/2010/08/call-for-benchmarks.html>
> 
> Hi, Lukas, that sounds like a pretty good project, both for Pypy and Bazaar.
> 
> The short answer is that fetching a large repository like lp:emacs or
> lp:launchpad and then running 'bzr pack' or committing a lot of
> changes is a decent test.
> 
> John Arbash Meinel and others have been doing some memory optimization
> work and you can find some descriptions by looking back through the
> list archives.
> 
> Another option is to look for bugs tagged 'memory'.
> 
> And finally the bzr usertest plugin has some macrobenchmarks.
> 
> Stay in touch,

The only problem with the test is that PyPy isn't compatible with our
Pyrex/C extensions. And while the pure python code always works, it
isn't 100% identical, and was certainly never tuned for memory or
performance. (It was tuned for readability, so that you have a reference
to crib from, etc.)

So while it is certainly true that doing "pypy bzr branch lp:launchpad"
is going to have very high memory consumption, it isn't necessarily
because of pypy.

John
=:->

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAkzzrswACgkQJdeBCYSNAAP0KgCfZit2ki9wLIQS2JI0v/LrJDcH
vRUAoMgz/iZFc3G5qSDJsjpySawb6SRq
=UzZm
-----END PGP SIGNATURE-----