[p4] backups

[G Barthelemy]

> What we plan to do now is to mirror our server to a virtual server in our
> virtual server farm. The cost for this is negligible (in terms of hardware
> that is) and even if a virtual server probably is not ideal for a p4 server
> we should be up and running within minutes if the server crashes and only
> loose the last 15 minutes of submits or so.

If that's of any reassurance, we are running our production Perforce
server (Linux) in a virtual environment (ESX server, DL585 hosts) with
all the storage (including the system disk) off a NetApp filer over
iSCSI (with a couple of Gb/s dedicated links for iSCSI). This works
well for us. We have over 600 Perforce users (but admittedly less than
that is using the server at any one time: we were averaging 8 to 10 p4
commands a second this morning for instance).
Incidentally we've had to "VMotion" the server (moving it from one
host to another while live) a couple of times in the last year, while
in use at peak time: the users didn't notice anything. So my
experience is quite different: virtualisation is great for a p4 server
(Maybe I'll sing a different song when we are 2000 users, but who
knows what the technology will be capable of  when we get to that
point ...)

Sticking to the original thread, having the DB, journal and RCS on a
filer opens up possibilities with regard to backup. Most filers have
now some sort of one-way mirroring capability which would allow
performing integrity check (p4 verify) and live backups (using a
separate p4d instance for checkpointing, maybe from another virtual
host) without having to replay journals to a cold database.

Guillaume