[darcs-devel] [patch1256] optimization: compress patch data before... (and 2 more)

[Ben Franksen]

Ben Franksen <benjamin.franksen at helmholtz-berlin.de> added the comment:

> Does block buffering mode improve speed?

I did not measure it but just assumed that block buffering must be
better in this case.

What a mistake!

I have now made a few test runs in which I explicitly set either no
buffering or block buffering in the local (Darcs.Util.External) and the
remote darcs (in Darcs.Util.ByteString.gzReadStdin) and ran all 4
combinations with both small and large pushes.

The result is that buffering seems to make no significant difference for
the sender (local darcs). On the receiver (remote darcs) no buffering
seems to give slightly better numbers; however, the difference is very
small and almost lost in the noise.

This is all completely contrary to my expectations.

> Is it necessary for the remote darcs to be in block buffering mode?

No. Both ends can independently chose buffering however they like.

> Is switching to block buffering mode an independent change from
> compressing the data?

Yes. I really should have recorded two separate patches. And I wish I
would have waited before pushing the patch to reviewed. Can we just
obliterate it (from screened and reviewed) and let me send an amended
version?

__________________________________
Darcs bug tracker <bugs at darcs.net>
<http://bugs.darcs.net/patch1256>
__________________________________