New benchmakt tests

Carl Friedrich Bolz cfbolz at gmx.de
Thu Aug 3 14:42:22 BST 2006


Robert Collins wrote:
> On Mon, 2006-07-24 at 09:28 +0200, Jan Balster wrote:

> And finally, the bundle tests are all not-quite there in two respects.
> Firstly they are testing the end user api - and we want bundles tested
> as a library tool, because the smart server will be using them in that
> manner.

Using write_bundle and read_bundle? Or anything else?

> Secondly, they are testing the creation of a bundle with only
> one revision - what we need is creating a bundle for some data points in
> a space which there are currently a number of interesting dimensions:
> Number of files in the bundle x number of files in the tree x number of
> revisions in the bundle x number of revisions in the tree.
>
> Of those four dimensions, the total number of revisions in the tree is
> probably not all that interesting *at the moment* - but is useful to
> benchmark because its a likely scaling dimension.
>
> So to parameterise this:
> file count in bundle:
>  bundle with few files (say 5), bundle with a number of files (say 100),
> bundle with many files (10K)
>
> file count in tree:
> small tree (say 5 files)
> moderate tree( say 1000 files)
> huge tree (10K files)
>
> number of revisions in bundle:
>  1 revision
>  500 revisions
>  10K revisions
>
> which is 27 (3*3*3) combinations. I think all the combinations are all
> interesting, even though creating a tree with 10K revisions and 10K
> files will take some time.

Ok, the above makes sense to, apart from the fact that not all 27
combinations are possible, as far as I see. How could you have more
files in the bundle than in the tree? So there are only 3*(1+2+3) = 18
combinations, which is still quite some and will probably take a long
time to run :-).

Cheers,

Carl Friedrich




More information about the bazaar mailing list