On Tue, Aug 24, 2010 at 18:39, dherring@tentpost.com wrote:
It has been my experience that synchronizing large projects is generally faster with git since it just grabs a few patches instead of checking each file individually. Git is only somewhat slower when doing the initial clone. That said, http-backed git repos are generally a few times slower than repos using the native protocol.
I'll chime in with some data about the slime git mirror I've been keeping:
The (packed, compressed) repo size is 79MB right now, and a fresh clone fetches all of it (unless you use --reference on an existing clone, which speeds things up a /lot/). This means a fresh checkout can take a while on a slower line. Just yesterday, I had to wait more than just a few minutes for my 2MBit/s line to finish sucking down slime. However, I don't expect hg or bzr to perform much better (or worse) than this.
However, an update from one day to the next generally takes ~1 second, transferring just a couple dozen kilobytes on very few server round trips (could just be one, I don't know the git protocol by heart). Again, I don't expect any modern VCS like hg or bzr to perform much better or worse than this.
Speaking as a user, I'd be happy to have the authoritative slime development history available offline, with uniquely identifiable commits, so any of the saner VCSs is fine by me.
Cheers,