[Systems] Hosting at RIT

[Bernie Innocenti]

So we have a deal for free hosting at RIT.

Ivan and I have been discussing hardware options over the fast few
weeks. We believe we have found a decent 8-core machine for
approximately $3000.  Ivan offered to pay a share of this sum.
I believe we have $2000 earmarked for infrastructure.

The use we plan to do for this new machine is:

1) replace our aging main machine solarsail, which runs
   www, wiki, trac, planet and mailing lists

2) running aslo, which is currently saturating the processor
   on sunjammer, but should have plenty of room for growth on
   this box

3) House some of the small virtual machines that are currently
   hosted in disparate locations

4) host a few unrelated domains previously hosted on solarsail

Point (4) may sound a bit controversial, but I think it's a fair deal,
considering that we've been using solarsail for as long as 2 years
without paying. All our infrastructure is currently running on third
party hardware usually shared by other services.

This not only reduces operating costs for us ($0 so far ;-), it is
primarily a way to maintain friendly work relationships with our
partners and sponsors. I ultimately believe that friendly agreements
with people we trust personally are worth more than any support
contract.

On the other hand, having our infrastructure fragmented in too many
disparate locations is costing us a lot in terms of sysadmin time, so
we're looking to consolidate our services on fewer physical machines in
2-3 locations.

It's not yet decided how much we'll rely on virtualization to partition
our hardware resources. There are good technical arguments both in favor
and against it. I'm not strongly opinionated either way, so I guess we
may decide what works best for us on a service-by-service basis.

Thoughts?

-- 
   // Bernie Innocenti - http://codewiz.org/
 \X/  Sugar Labs       - http://sugarlabs.org/

-------- Forwarded Message --------
From: Paul Mezzanini <pfmeec at rit.edu>
To: Bernie Innocenti <bernie at codewiz.org>
Cc: Stephen Jacobs <sxjics at rit.edu>, David Farning
<dfarning at sugarlabs.org>, Charles J Gruener <cjg9411 at rit.edu>
Subject: Re: Sugar Labs hosting at RIT
Date: Wed, 16 Dec 2009 12:06:14 -0500

Charles and I had a good long talk about what hardware is available
and its capabilities.  We are both in agreement that the hardware
available for donation is inadequate for your needs.  

What we propose is:
Sugarlabs buys the primary server and ships to RC.
RC racks / physically installs the machine and hands IPMI access to
sugarlabs for final config.
RC adds the primary server to its backup system for nightly backups.
RC provides a passive failover server located in RCs virtual
infrastructure.  

How does this sound?  Final details can be hashed out after we get the
groundwork set.  

-paul


On Dec 16, 2009, at 11:11 AM, Bernie Innocenti wrote:

> On Tue, 2009-12-15 at 22:14 -0500, Paul Mezzanini wrote:
>> I need to run it by my boss. How much bandwidth do you estimate you

>> use?  I need to see if I need to get approval from the networking
>> team
> 
> Currently, we serve ~1000 reqs/sec at peak time, using up to
> ~15Mbit/s for a few hours in the morning. We estimate to double
> or triple these numbers within 6-12 months.
> 
> The actual content is already being off-loaded to a CDN. We may
> further reduce bandwidth by offloading static files to machines
> hosted by the FSF or MIT.
> 
> 
>> I may be able to provide a backup machine in ESX for failover too.
>> There are other hardware options available depending on if you need

>> full virt or para virt. Also on how much io you really need. 
>> Charles (cc'd on this) can chime in on exactly what he has.
> 
> As long as performance does not suffer, I have no preference.
> I guess full virtualization leaves us a little more flexibility in
> updating the kernel.
> 
> We do limited I/O at this time because our workload is read-mostly
> and we have a lot of RAM to cache disks. We're saturating the CPU
> much before we start seeing any disk trashing.
> 
> -- 
>   // Bernie Innocenti - http://codewiz.org/
> \X/  Sugar Labs       - http://sugarlabs.org/