[MERGE] first round of use case benchmarks

[Carl Friedrich Bolz]

Sorry for the slow response (was busy writing a paper for uni).

John Arbash Meinel wrote:
> Carl Friedrich Bolz wrote:
>> John Arbash Meinel wrote:
>> [snip]
>>> The permutations seem reasonable, though.
>>> I wonder if it would be better to separate the timing. 1 time evaluated
>>> for generating the bundle, and another time for reading and installing
>>> the bundle. Though that ends up doubling the number of tests you write.
>> Hum. I agree in principle, but it seems to just take too long to build
>> the trees. The test_many_files_big_tree_1000_revision takes a _long_
>> time on my machine just as it is, so building the tree twice does not
>> sound like it is a good idea to me.
>
> Actually, that is probably a very good idea. We can have separate tests
> for up to maybe 100, and then a big 1000 commit test, that just does
> both. In the future maybe we can break it up, but in the short term,
> that is what the small tests are for.

It seems the 1000 commit test is just too slow when writing and
installing the bundle (2.5 hours, time spent mostly with installing the
bundle). So what I did was to split up the tests but not run the 1000
revision tests with install_bundle.

[snip]
>> +    def test_few_files_small_tree_500_revision(self):
>> +        os.mkdir("a")
>> +        tree, files = self.create_with_commits(5, 1, directory_name="a")
>> +        self.commit_some_revisions(tree, files[:5], 500, 1)
>> +        self._time_read_write()
>> +
>> +    def test_few_files_small_tree_1000_revision(self):
>> +        os.mkdir("a")
>> +        tree, files = self.create_with_commits(5, 1, directory_name="a")
>> +        self.commit_some_revisions(tree, files[:5], 1000, 1)
>> +        self._time_read_write()
>
> It occurs to me that doing 1, 100, 1000 might be a better distribution.
> And we might even want to wait on the 1000 for now, if they really are
> as bad as they seem. (Or maybe just make an easy way to avoid running
> them with the rest of the tests).

Did this. The branch is registered at launchpad as:

https://launchpad.net/people/cfbolz/+branch/bzr/usecases-benchmarks

Cheers,

Carl Friedrich

-------------- next part --------------
A non-text attachment was scrubbed...
Name: bench_usecases.patch
Type: text/x-patch
Size: 20558 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20060907/f0e84e45/attachment.bin