On tis, 2008-07-15 at 23:42 +1000, Robert Collins wrote: > On Tue, 2008-07-15 at 23:37 +1200, Amos Jeffries wrote: > > For the record. Robert your method of merge did eventually work. It > > seems I was just too impatient with my first attempt. > > > > The URL merge from launchpad just took me an hour to complete. Before > > any of the testing began. > > That sounds funny; what commands did you use to do this? (it should only > have had a few hundred K to download to work ...)
The readv over http one at a time with no pipelining gets quite slow when there is even a little bit of latency. It's not the bandwidth being the main bottleneck for bzr operations, except maybe when one is doing an initial pull of a repository without using a local shared repository already having most of the data.. I do have a feeling the bzr index reading procedure for finding the common parent has a major error somewhere, probably wandering off reading lots of random remote stuff it doesn't really need. Seems to need way too many roundtrips even when there is only very small number of changesets to query for.. A thought is that perhaps it has got the remote/local site of the equation wrong.. Regards Henrik
