On Tue, Jun 16, 2009 at 9:10 PM, Daniel Kinzler <daniel.kinzler@wikimedia.de> wrote:

> Total disk resources? (where are the server specs anyway?).

Looking at the specs again, I see I was wrong - the toolserver is going to have
far less room than that. It really can't take 350 GB, it only 584 total. I was
calculating with 2TB, but that's the DB server.

Reiterating the server specs:

* Cassini, the Toolserver: X4150, 2x4 cores, 24 GB RAM, 4*146 GB Disks ~ 1/2 TB
* Ptolemy, the database: X4250, 2x4 cores, 32 GB RAM, 16*146 GB Disks ~ 2 TB
* Ortelius, squid/render: 1x4 cores, 28 GB RAM, 4*146 GB Disk ~ 1/2 TB

> XAPI is useful, and if we can host it we probably should, it'll be
> useful to some toolserver users and we can always throw it out later
> if it becomes a problem.

So, this means it can only be on the production servers. That would mean
extending the original profile the servers were layed out for, and it raises the
question of access. How is the XAPI accessed? Is it a standalone server? Does it
need apache? So far, I think Ortelius would run squid and lighttpd.

XAPI is pretty independent of anything else. The api is published via http. Usually using Apache. I've not tried it with lighttpd but it only requires CGI and URL-rewriting. lighttpd supports both of these so that shouldn't be an issue.

Also, who would maintain it on the production boxes? And is mark OK with having
an extra service there?

I'd be prepared to maintain it, wherever it ends up.

80n