[darcs-users] darcs2 performance regression

[Jason Dagit]

On Wed, Apr 30, 2008 at 9:37 AM, Simon Marlow <simonmarhaskell at gmail.com>
wrote:

> I've switched to using darcs2 locally (no repo changes yet, just to see
> how well it works as a drop-in replacement).  It seems to be noticeably
> slower for many of the operations I do day to day.


You listed whatsnew and amend-record.  Do any others come to mind?  Have you
enabled the global cache mechanism?  Have you experimented with the
repository format upgrades?  So many questions, so little time :)

Although switching to darcs-2 format requires a centralized conversion you
could start using hashed format locally without impacting anyone else.  I
think you might see some improvements by doing so.

Details about global cache:
http://darcs.net/manual/node5.html#SECTION00510070000000000000

Details about hashed format:
http://darcs.net/manual/node7.html#SECTION00742000000000000000
(look at get --hashed)


>
> Here's an example, doing 'darcs whatsnew' on the GHC testsuite
> repository (http://darcs.haskell.org/testsuite):
>
> $ time /usr/bin/darcs w -s
> M ./tests/ghc-regress/ghc-e/should_run/Makefile +3
> M ./tests/ghc-regress/ghc-e/should_run/all.T -1 +1
> A ./tests/ghc-regress/ghc-e/should_run/ghc-e006.hs
> A ./tests/ghc-regress/ghc-e/should_run/ghc-e006.stdout
> 0.97s real   0.24s user   0.09s system   34% /usr/bin/darcs w -s
> $ time darcs2 w -s
> M ./tests/ghc-regress/ghc-e/should_run/Makefile +3
> M ./tests/ghc-regress/ghc-e/should_run/all.T -1 +1
> A ./tests/ghc-regress/ghc-e/should_run/ghc-e006.hs
> A ./tests/ghc-regress/ghc-e/should_run/ghc-e006.stdout
> 8.03s real   6.98s user   0.46s system   92% darcs w -s
>
> 8x slower!  (and no progress messages)


Interesting.  You're not the only one to mention that 'whatsnew -s' is slow
and there has been a buzz about it.  Recently we found a way to make the
binary file test quite a bit faster, which may impact your use case for
whatsnew.  I was under the impression darcs2 was actually faster during
whatsnew because of the hashed pristine, but it sounds like it's actually
worse.

I'm fetching your repository now so I can run 'darcs show repo' to find out
specifically which repository format options you have going on but it will
take me quite a while to fetch it.

yesterday I noticed that an amend-record in the GHC repo was quite slow,
> and the progress message reported that it was reading the inventory.
>
> I hope this is of use to someone...


Always useful to get feedback.

Thanks,
Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.osuosl.org/pipermail/darcs-users/attachments/20080430/82c90dac/attachment-0001.htm