[Server-devel] location and muscle of the school server

Sameer Verma sverma at sfsu.edu
Fri Feb 22 02:20:11 EST 2008


John Watlington wrote:
>
> On Feb 19, 2008, at 9:28 PM, Sameer Verma wrote:
>
>> I've been reading up on the specs for running the XS (XSX - the one that
>> relies on off-the-shelf parts) and one thing that keeps bugging me is
>> that if we don't have reliable power infrastructure on site (which is
>> why XO makes a lot of sense) how can we expect to run a
>> common-off-the-shelf server that consumes 200W of power?
>
> We don't.   Unfortunately, real world resource constraints impose
> themselves --- OLPC doesn't have the resources to develop the
> server hardware until later this year.   In the meantime, software
> development can proceed using available hardware.
>

OK, but the constraints will change once the XS hardware becomes 
available. With the XO, the development was on prototypes (ATest boards 
and onward) that led to the MP machines. All along, the constraints have 
grown with development. However, that's not the case with XS. When the 
XS machines do arrive, Will the cat fit in the bag at that point?

> The off-the-shelf hardware that I'm currently testing uses 18 to 22W,
> including the disk drive and three active antenna.
> Not quite the 200W you cite...
>
Wow! Most of the stuff I have will do at least 100W+. Do you have any 
details on your setup?
> There are schools that do have power, and are large/dense enough that
> they need a traditional access point deployment.   These schools
> WILL deploy with off-the-shelf hardware.
>
> The reason to build an OLPC server is mainly environmental and cost.
> Any server designed to run in harsh environments (temperature, dust,
> and corrosion) is currently very expensive.   We can also drop the
> power dissipation even more by moving away from x86.
>
>> I see in the specs that the model of using an XO with external drive as
>> a school server is an option, but I wonder how far that has been
>> explored. It is a great idea (after all, all nodes on a mesh are just
>> that - nodes) but given the current development, it doesn't look like an
>> XO has enough to run Squid, etc. I take it that the current run of XS is
>> short-term?
>
> The XO doesn't have enough memory to be a school server, and
> the external disk drive needed is difficult to power from the internal 
> battery.
> We are still exploring our options here.
>
>> Another point that came up a couple of days ago (when I was discussing
>> this with a colleague) is why can't the server be further upstream at a
>> data center where it has adequate power and cooling?
>
> Because less schools have internet access than power.  And if they
> do have internet access, it is a few Mbits/sec.    Local storage is
> critical to providing a decent user experience.
>
> Think about you just proposed.   You want a hundred students to all
> simultaneously try to download a 2 MB PDF over a 2 Mb/s line ?
>
Won't caching solve that to some extent?
> Yes, there will be servers farther upstream, in nice cooled rooms,
> providing additional storage backup and content/applications.  But
> content used by the school must be cached locally to minimize
> access latency.
>
Can caching be accomplished on low power XO like machines?

Sameer

-- 
Dr. Sameer Verma, Ph.D.
Associate Professor of Information Systems
San Francisco State University
San Francisco CA 94132 USA
http://verma.sfsu.edu/
http://opensource.sfsu.edu/



More information about the Server-devel mailing list