[PATCH] support streaming clone

Vadim Gelfer vadim.gelfer at gmail.com
Sat Jul 15 05:02:19 UTC 2006


On 7/14/06, Matt Mackall <mpm at selenic.com> wrote:

> So you've found a bug that needs fixing. There's no reason that clone/pull
> footprint should be bigger than:
>
>  O(number of files + largest file revision)

i will try to reproduce problem.

> Eh? If we have a 10Mbps pipe and I can do a clone that saturates the
> pipe with 100MB of data, to have 10x better performance, you'd have to
> do it with 10MB of data.

i mean local network. my local network is 100Mbps and 1Gbps. pull is
100% cpu limited there. stream is 100% io bound. only local network i
have that is 10Mbps is wifi.

> Under 10Mbps, unrecompressed transfers are going to be 40% slower than
> pull.

yes. that is same as what bos measured. but i want to be fast on lan
and wan, so i want option to stream for lan.



More information about the Mercurial mailing list