[Systems] Clustering Remora

[David Farning]

On Thu, Aug 27, 2009 at 9:09 PM, Bernie Innocenti<bernie at codewiz.org> wrote:
> El Thu, 27-08-2009 a las 19:59 +0200, Marten Vijn escribió:
>> As posted some ago download2 does not always have a .stamp file.
>>
>> Currenty I am decreasing verbosity.
>
> I had no time to investigate yet.  Do you already have a shell account
> on sunjammer?  And would you like to have a look yourself?
>
>
>> Funny see my posting on cacheboy asked for bandwidth. Getting this for
>> them would also help us distributing iso's.
>
> Answered that.  Let's see what we can get from the GNAPs/MXP folks.
>
>
>> Think getting more data from trend and monitoring i next important step
>>
>> Also moving my mirror_check script should be moved to a monitoring
>> machine.
>
> Are you missing a VM to do this?  Me and Dogi are setting up a new KVM
> host that should be ready for production within days.
>
> If you'd prefer to use your own resources, go on and do what you think
> would work best.  Only, please also document what you do in the wiki and
> give access to a few more Infrastructure Team people.  As we grow
> bigger, I think it's important that we transition from being a bunch of
> individuals to a coordinated team of interchangeable players.
>
>
>> This could help finding bottle neck. At some point we could need a setup
>> with loadbalancers / failover mechs, but I don't think we are there yet.
>
> At this point, bandwidth is not /yet/ a big bottleneck for us.  It will
> certainly become a problem at some point, so it's good to start planning
> ahead.  Besides, regardless of how much bandwidth we have at the source,
> offering geo-located mirrors can always be a good service to users
> located far away from the main distribution site.
>
> My #1 concern to scale software distribution would be properly
> clustering a.sl.o.  Remora is a fairly large and slow LAMP application,
> and no matter how much we can optimize away inefficiencies, we'll
> eventually reach a point where no individual machine will be able to
> cope with the traffic.

If the resources are available, the first piece to split off would be
the file server portion of a.sl.o. That is just a matter of setting up
a very light weight http server on a machine with good bandwidth and
adjusting a single line of the a.sl.o config file.

> Unfortunately, I'm afraid the application servers will have to sit near
> the only database server, or performance will suck.  Perhaps Remora
> supports a master-slave MySQL scenario, I haven't investigated.
>
> This would also be another good project waiting for a volunteer sysadmin
> to explore.  I can provide all the necessary resources and support.
> activities-devel.sl.o. would be a good playing ground, just ask me to
> get access and get in contact with Aleksey for coordination.

Hopefully, that is something wes is able to do.

How about moving activites-[devel|testing].sl.o to bender or treehouse
in order to move them from tier 1 to tier 2 machines?

david

>   // Bernie Innocenti - http://codewiz.org/
>  \X/  Sugar Labs       - http://sugarlabs.org/
>
> _______________________________________________
> Systems mailing list
> Systems at lists.sugarlabs.org
> http://lists.sugarlabs.org/listinfo/systems
>