brisbane-core+ric-commit results

John Arbash Meinel john at arbash-meinel.com
Thu Mar 26 21:59:15 GMT 2009


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robert Collins wrote:
> On Thu, 2009-03-26 at 12:43 -0500, John Arbash Meinel wrote:
>> I was hoping that Robert or Andrew could help with changing the fetch
>> code, since it integrates deeply with their streaming work. Which
>> isn't
>> something that I fully grasp.
> 
> Sure thing.
> 
>> How would you define a new network_name() so that you can request data
>> be streamed in 'groupcompress' order, etc. I've already written a
>> rather
>> decent optimized streaming in GCPacker (which does revs, inventories,
>> chk pages in a good layout, and then texts). I just don't know how to
>> hook that in to Fetch.
> 
> Just set format._fetch_order.
> 
> -Rob

So it is really more about getting a different StreamSource. The one we
have still double-walks the inventories. And that is significantly more
expensive for CHK repos (though my latest commits have made it better).

I don't really care about streaming in 'groupcompress' order, because if
you do "bzr pack" it will rewrite your content into groupcompress order,
such that 'unordered' is *still* groupcompress order (because it uses IO
order).

Without changing the btree leaf-node cache size we end up spending 458s
in 'iter_interesting_nodes', so if we only did it once we would save
3.8min during 'bzr branch'. (If we increase the btree cache size, we
only spend 400s in iter_interesting, so we would save 3.3m.)

John
=:->
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iEYEARECAAYFAknL+rMACgkQJdeBCYSNAAP5RACgkgCVMxT4/3wP7etMtF5xxlen
4roAoLqH5RovukQOeVLSOZc2if0C1ouh
=+itQ
-----END PGP SIGNATURE-----



More information about the bazaar mailing list