On Wed, Nov 19, 2008 at 7:44 AM, Eric Kow <[EMAIL PROTECTED]> wrote: > Hi, > > On Tue, Nov 18, 2008 at 21:49:37 -0600, Tom Hawkins wrote: >> We use darcs at work; and it works great. > > Thanks for the encouragement. > >> The one exception is it's slow. We are still on 1.0.9 due to cygwin >> support, so maybe things have greatly improved. > > I'll refer you to Zooko's response > >> But in case they haven't, I'm wondering if >> anyone has considered using digital fountain codes as a means to >> communicate patches between repositories? > > With apologies for being obtuse, this is something you are proposing as > a way of speeding up operations like darcs get and pull. In other > words, this is proposed as an optimisation? My reading of the below was > that we could use this as a way to simultaneously fetch different pieces > of the same patch(es) from more than place. Sorry for trying to > executive-summarise it :-)
It occurred to me yesterday when explaining the darcs cache to a friend that we could potentially get a speed up in the cache by fetching from the various repository sources in parallel. Unlike adopting Fountain Codes this is an optimization that could probably be added in a day given our existing framework and the fact that we already use threading. Plus, as I understand it, it would help with the bug Zooko reported in the last month where our approach of going down the sources list sequentially means that sources which timeout make certain operations painfully slow even though at least one of the sources was available with the data. The clever trick you'd want to add is that once you start transferring a patch from a source you want to cancel the request of that patch from the other sources. Why? Because I think otherwise you'd lower through put of the patch you want by sharing the bandwidth with extra copies of it. Granted, Fountain Codes wouldn't have this problem as I understand it. We could also consider distributing the request list over the sources (again in parallel) and if a request fails at a particular source we try it again from a different source. Well, just some thoughts. Jason _______________________________________________ darcs-users mailing list [email protected] http://lists.osuosl.org/mailman/listinfo/darcs-users
