Re: User services (was: Re: dict server)

[Owen Taylor <otaylor redhat com>]

On Thu, 2006-04-06 at 13:51 +0200, Reinout van Schouwen wrote:
> Hello Owen, all,
> 
> On Mon, 30 Jan 2006 17:51:19 -0500, Owen Taylor wrote:
> 
> Sorry for getting back on this so late, I'm swamped with my thesis and
> had forgotten about it until now. To refresh everyone's memory, this
> discussion resulted out of bug 167366, it was a request to run a
> dictionary server on gnome.org.
> 
> > I don't think we need to be scared of the idea that we might swamp our
> > limited server capacity (even 8 servers is a pretty small set), or swamp
> > Red Hat's bandwidth. If we create things that are genuinely useful and
> > interesting to people, it's easy to fund-raise for new servers, and it's
> > easy to find people to host bandwidth. Swamping is good.
> 
> So if I understand correctly, Red Hat is willing to add more hardware
> and bandwidth if the services we want to offer require it. 

I'm not sure that's quite a correct reading of my email :-)

What I'm saying is that if people can create services that excite people
and cause big server demands, than the GNOME foundation can fundraise
and/or find donors to meet those demands. Red Hat is one possible source
of resources, but certainly not the only one.

I think it makes a lot of sense to host services in a developmental
stage on the main GNOME cluster at Red Hat, but we have to be able to
be flexible about being able to offload if we get huge demands.

> The question
> then is, how do we move forward on this issue? Who should we talk to about
> adding this service? It would be nice to have it ready in time for testing
> during the GNOME 2.16 release cycle.

In terms of the dictionary server in particular, I think some of the
questions that the sysadmin team would be interested in a proposal
would be:

 - What the specific benefit to the users from this server?

 - What is the best guess we have at this time about the load on
   the servers ... how many requests a day, how much bandwidth,
   how much memory usage from the server?

 - What's a worst-case estimate as to the above?

 - If the servers were getting swamped:

   - Could we repoint clients to another machine without updating
     the software?
   - Can the service be distributed onto multiple machines?

 - What would be the effect on people if the service went down?
   Is that handled gracefully?  What are the necessary reliability
   guarantees?

 - If a distribution wanted to configure what they were shipping
   to point elsewhere, could they do it?

   (There really needs to be some central place where all network
   resources referenced by components of the GNOME software 
   are documented, with information about the effect of firewalls,
   how to repoint them, etc.)

Assuming that the projected demands aren't too heavy, it's quite likely
we (GNOME) could host something on one of our machines, at least
assuming that the current plans to get a couple of more machines online
in the near future work out. 

					Owen