[cvsnt] sync-protocol questions

[Gerhard Fiedler]

Arthur Barrett wrote:

>> I disagree. The main (only?) advantage local repo replication brings is
>> fast arbitrary diffs, and I think there is no other way to get this
>> than to locally replicate.
> 
> On even a very slow network a diff could take no longer over the WAN
> than locally.  When we switched to duplicated repositories in Sydney and
> Manchester I noticed absolutely NO difference in the speed of a diff and
> the overall speed of our link is little better than dial-up.

Hm... strange. Note that I'm talking about the TortoiseCVS/WinCvs external
diffs which fetch one or two full revisions of a file into local temp files
and diff them with a GUI differ. 

I'm working mostly with one local CVSNT server (a really old and slow
machine) and one remote server (a much faster machine, through a SonicWALL
VPN). Local diffs in the LAN of the remote server are quite fast, and local
diffs here with my own, slow server are still fast, but remote diffs from
here to the remote server are noticeable slower (about twice, and I think
the difference between local diffs in the remote LAN and remote diffs is
still more, because that server is faster).

When I have time I get some hard numbers for various scenarios, but it
seems obvious to me that running a diff (effectively a checkout of one or
two files, typically some 50k or so) is snappier on a LAN than over a
remote connection. Diffing two arbitrary revisions of a 50k file takes
about 4s on my slow local server and 8s on the fast remote server.

> If your diffs are taking a long time you need to follow the same
> procedure I outlined for Carsten - investigate why the diffs are slow
> (both technical and procedural).

Well, there's not much to the procedural part. Of course I have separate
checkouts for interesting branches and tags so that I can diff locally over
whole directory trees when needed, but there's still the need to
occasionally do arbitrary diffs between arbitrary revisions of a file.
That's just how life is :)

As I said, I'll run some tests, but it seems to me that what I see is just
plain normal for working over a remote pipe. 

Thanks,
Gerhard