bzr wins a performance test

[Jonathan S. Shapiro]

On Wed, 2008-01-23 at 13:18 -0500, Theodore Tso wrote:
> So it's very hard to make a comparison, and basing it just on space is
> clearly insufficient.

Yes. One thing does seem pretty clear though: one FS file per SCM file
is probably okay, but one FS file per internal SCM schema object is
probably not. For example, the OpenCM schema has something called an
Entity that occupies about 128 bytes. Spending 4k to 8k on that (ext3
block size) added a truly astonishing overhead.

Mind you, the "one file per object" in OpenCM was for convenience of
initial implementation. We certainly knew about this issue, and we
certainly intended to resolve it by moving to a more sophisticated store
later. It was a fine first implementation.

And educational. I say "we knew", but the visceral impact of seeing the
numbers was still pretty astonishing.

shap