OLPC/P2P

[Jim Gettys]

Note that backhaul bandwidth to the internet will often be precious in
our installations.

So it is very important that any file sharing/caching system be aware of
this, and try hard to find its data "locally", if at all possible.  We
really only want to have data transferred over the long haul link (often
satellite) from the internet only once.

Part of the role of the school servers is to mediate this access, and
allow for much larger scale caching of content (as it will have disk
drives where significant amounts of content can be cached).
                                         Regards,
                                                 - Jim


On Mon, 2007-03-05 at 12:16 +0100, Michel Meulpolder wrote:
> Hello SJ,
> 
> > Next steps: we could use specific use cases.  What are the use cases
> > for which Tribler and P2P-Fusion are designed, what are  
> > expectations of network connectivity, what are expectations of  
> > availability of material?
> >
> 
> I think for OLPC the core of the Tribler technology is most relevant.  
> The core focusses on exploiting the available resources as  
> efficiently as possible, with highest possible performance. This goal  
> is basically the same for all connectivity/availability scenarios:  
> getting the most out of what's available. On top of this technology,  
> various Tribler and Fusion applications are built which focus on  
> modern online social-aware users in many different contexts. In my  
> view the core technology can perfectly be used for new applications  
> that suit the specific conditions/expectations of OLPC target use. We  
> can take OLPC use cases as a focus and see how Tribler technology can  
> be customized.
> 
> Regarding availability: One of my research-topics is automatic  
> replication to make sure that content is always available in the case  
> of sporadic connections. Peers can add bandwidth for specific content  
> based on real-time popularity statistics.
> 
> > One problem we need to solve is how to discover material on a  
> > sporadically connected network, and how to efficiently track where  
> > such material might be found.  The interlibrary loan system and  
> > Tribler handle this in different ways; we may want something  
> > different from either.
> 
> The tracking problem itself is in current systems still done with a  
> central component (a tracker like in BitTorrent). Online peers  
> automatically notify the tracker with what they have available. The  
> metadata contains tracker-addresses, so that the requesting peer  
> automatically asks the tracker for addresses of peers that share the  
> required content. In practice, this setup works very well, and  
> trackers can easily be set up for both small and large situations  
> (ranging from private LAN to internet-wide). Due to the very small  
> amount of traffic for tracking information, such a tracker would  
> become a bottleneck only in very extreme and unlikely cases. However,  
> we are currently working on fully distributed tracking/discovery  
> protocols as well.
> 
> - Michel
> 
> >
> >
> > On Mon, 5 Mar 2007, Michel Meulpolder wrote:
> >
> >> Hello SJ and Ivan,
> >>
> >> A short follow-up on our mail of last month. I am very interested  
> >> in all P2P developments in OLPC and curating relevant projects.  
> >> There are many interesting developments here by the way in P2P/ 
> >> Tribler and also in the European P2P-Fusion project that I am  
> >> involved in (concerning development of a large P2P mediaspace for  
> >> open and closed communities).
> >>
> >> My question for now: What is the next step? Just let me know how I  
> >> can get involved / what activities are relevant.
> >>
> >> Regards from the Netherlands!
> >> Michel
> >>>
> >>>> I see some serious options here, since the Content  
> >>>> Distribution / performance issue is exactly what I have  
> >>>> specialized in during my M.Sc. research, and what I would like  
> >>>> to continue. The idea that I focussed on until now is to equip  
> >>>> end-user machines and school servers with Tribler/BitTorrent- 
> >>>> powered P2P caches, transparent to the user. With these  
> >>>> protocols, it is possible to transmit large amounts of content  
> >>>> to thousands of users simultaneously without decreasing speed,  
> >>>> since it upscales with the number of users. Last year, I have  
> >>>> designed and implemented a prototype version of this software  
> >>>> and tested it in Manila at a large campus, and the results  
> >>>> showed a tremendous potential. I am fully convinced that a grown- 
> >>>> up version of this solution would solve the performance/caching  
> >>>> issues mentioned in http://dev.laptop.org/ticket/15. Longer-term  
> >>>> developments could include in-browser P2P video-on-demand and  
> >>>> P2P webhosting, which are hot topics at the research department  
> >>>> here. In essence, the software and protocol could be the ideal  
> >>>> "multicast file transfer protocol" that is talked about in the  
> >>>> ticket.
> >>>> Are there already developments in this particular direction?  
> >>>> What kind of specific P2P protocols/applications are used in the  
> >>>> current laptop implementation? In my current scans in the  
> >>>> developers archives I did not yet find a lot of P2P-related  
> >>>> topics, but perhaps I was not on the right track yet :-)
> >>>> Since the Manila project, several parties in SEA countries have  
> >>>> become very enthusiastic to be part of further deployment of the  
> >>>> software. This could be a perfect testbed and step towards  
> >>>> inclusion in OLPC. Please let me know what you think, I see a  
> >>>> major opportunity here.
> 
> _______________________________________________
> Library mailing list
> Library at laptop.org
> http://mailman.laptop.org/mailman/listinfo/library
-- 
Jim Gettys
One Laptop Per Child