smart server push

Robert Collins robertc at robertcollins.net
Mon Jan 7 02:15:02 GMT 2008


Hi Andrew, A little birdy told me that this is getting close to the top
of your queue :).

I had some thoughts about it that I thought I'd brain dump - please feel
free to ignore or not as you wish.

Firstly, I'm really keen to see this come about. I have a new inventory
format I will be putting up for review in a few branches shortly. It
changes both push and pull in that you no longer have a 'full text' for
an inventory, rather you have a 'journal' which is composed of many
'journal entries', all of which must be present to parse an inventory
for use. So I'm hoping this won't be too disruptive for smart server
stuff; but because it may be I'm focusing on getting it done asap.

Secondly, I think push is a good operation to 'go deep on' - to get into
a loop of
. benchmark
. analysis
. code (and send branch up ;))
iteratively.

I think doing this with packs is essential (and with development1
if/when that gets approval). 

I think some interesting benchmark dimensions are:
. *both* sftp and bzr+ssh because sftp's file_stream is pretty good,
packs should be better on sftp today.
. file count because scaling linearly is only ok if latency is not a
multiplier here:)
. revision count for the same reason

So you might like to grab e.g. a conversion of mozilla or open office
and use that to benchmark.

These are so big that simple wire transfer will hurt, so I'd definately
want to be benchmarking with near infinite bandwidth, but significant
latency. There's a number of tools for linux that will let you do that
on your laptop.

end-dump

-Rob


-- 
GPG key available at: <http://www.robertcollins.net/keys.txt>.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : https://lists.ubuntu.com/archives/bazaar/attachments/20080107/3d20692f/attachment-0001.pgp 


More information about the bazaar mailing list