[Server-devel] [IIAB] Internet-in-a-Box speed profiling, tips on different CPUs?

Tony Anderson tony_anderson at usa.net
Mon Jun 9 00:35:24 EDT 2014


I think the focus needs to be on the loads placed on the server by the 
workloads, not which hardware performs better. In general, server 
performance is a very interesting and challenging topic. The important 
point is that performance is limited to the speed of the slowest element 
in the process. This could be network speed, cpu speed, cache misses 
(cpu or memory), disk accesses and so on.

Google performs fast searches primarily be pre-indexing the content. 
Most Google searches do not
result in any links outside the Google system.

It could turn out that, once we understand how the server is handling a 
workload, is to change the way it is handled to something that gives 
servers better capacity. For example, suppose we provide a Pustakalaya 
style guide to books in Gutenberg so that most children pick a book from 
a set of 25 shown on a pre-built webpage.

Perhaps we discover that many WIkipedia searches are repetitive since a 
class is working on a common project. We might also discover that the 
squid algorithm is not caching the search results so
that they have to be repeated. Perhaps we then could come up with an 
algorithm that creates an dynamic index to be looked at first before a 
full search.

In the case of Open Street Maps, I suspect the work load will focus on 
network bandwidth in supplying tiles as the child moves their map on the 
screen or changes the zoom level. The reality is that we
can't speed this up by a compression scheme, for example, because that 
would increase the computational load on the XO. Maybe we come up with a 
strategy of a classroom level Raspberry Pi that can cache map tiles in 
some way assuming that most children are looking at the same region of 
the map.

Tony

On 06/09/2014 04:45 AM, server-devel-request at lists.laptop.org wrote:
> Message: 7 Date: Sun, 08 Jun 2014 15:46:40 -0700 From: Braddock 
> <braddock at braddock.com> To: Internet In a Box Working Group 
> <iiab at sgvhak.org>, server-devel <server-devel at lists.laptop.org>, 
> "Unleash Kids!" <unleashkids at googlegroups.com> Subject: Re: 
> [Server-devel] [IIAB] Internet-in-a-Box speed profiling tips on 
> different CPUs? Message-ID: <5394E7D0.2080907 at braddock.com> 
> Content-Type: text/plain; charset=ISO-8859-1 -----BEGIN PGP SIGNED 
> MESSAGE----- Hash: SHA1 On 06/06/2014 11:18 PM, Adam Holt wrote:
>> >As RichardS & GeorgeH valiantly get up to speed clocking
>> >Internet-in-a-Box bottlenecks on side-by-side school servers like
>> >Nepal's MSI DC111 (Celeron 1.8GHz, 2GB RAM) and faster, with many
>> >developing world deployments worldwide itching to follow -- who
>> >has intuition what they should test/compare first?
> The most intense Internet-in-a-Box features are Gutenberg search and
> wikipedia search.  Those are probably the performance pain points in
> scalability.
>
> - -braddock



More information about the Server-devel mailing list