Draft 0.1 2007-04-11
Draft 0.2 2007-05-05 see also bits in olpc thread


1. This is basically a collection of links and cryptic notes for possible future elaboration and separation.

2. I cannot work on this stuff at the moment and am basically incommunicado until I can work on things but points are numbered in case anyone wants to notify me of interest in elaboration of specific numbered points (via the personal message facility at this site).

3. OLPC has brilliant core principles and HW and software platform. Same platform and variants will become widely available for many other purposes. A key feature is the Sugar based User Interface (UI) which will presumably support interactivity through the wifi mesh using dbus over event based twisted framework with dynamic presence management and service discovery via Avahi multi-cast DNS. I have also seen reference to use of CSpace and presumably anyone capable of doing any of this sort of work would also be aware of the python spread module for group communications. The intention is clearly (and necessarily) to support interactive text and voice (VOIP) and video chat together with all other activities (including music with orchestra collaboration and jam sessions as well as just sharing). Telepathy provides basic mechanisms combined with dbus to handle lots of this stuff via increasingly sophisticated XMPP protocols (jabber). Not as independent of school servers as it should be.

4. OLPC model is for rollouts (initially via education Ministries), to saturated age cohorts at schools. This enables interactivity in group activities via wireless mesh both at school and in neighborhood at home.

5. Low power design and lower power System On Chip wifi that continues relaying wifi traffic when the main laptop is powered off combines with age cohort saturation to enable a village mesh around a school because one can expect there to be mesh nodes sufficiently close together.

6. In practice this is likely to rely heavily on daily battery charges at schools either connected to mains electricity power grid or with local school generator. But it does enable use at homes that have neither, and can be combined with (significant additional cost) low power generators (eg foot pedals, solar etc) for use pretty much anywhere that cell phones could be useful.

7. That is, pretty well anywhere:

A lack of electricity has not proved a hindrance: roadside vendors charge mobile phones with car batteries. As the signal coverage expands, cheaper phones and calls fuel growth.

Last month Safaricom started selling what is believed to be the cheapest mobile phone. Designed by Motorola for the developing world, it costs £20, with free connection. The cheapest airtime voucher is about 40p.

However it is either already well understood by the developers, or will quickly become obvious through trial deployment experience, that mesh connectivity at home (unlike at school) will be inherantly unreliable, even in places where homes are connected to the power grid and laptops can remain fully powered relaying the mesh indefinately via ordinary mains power adaptors. That is a logical consequence of the nodes being portable instead of fixed and prepositioned for coverage. In any case, interactive text and voice chat will only be possible when all relevant parties are present. When they are not, people will want to send text and voice messages for delivery when the addressee is present. That will work fine via store and forward relay no matter how unreliable the mesh is, since the kids take their laptops to and from school etc. Also if holding the E-book mode vertically antennas would be horizontally polarized instead of vertically and possibly pointing at each other's blind spots. Could require more complex swivel mount to fix although can be mitigated by user understanding in the same way as consequences of fully shut down or leaving uncharged on ability of others to communicate. So the UI will have to be designed to more or less transparently provide a powerful interpersonal messaging (IPM) facility easily used by 6-12 years olds, without support from local system administrators and help desks, including all the infrastructure such as address books, vcards and directory services that are necessary to enable this. I am not clear on the state of development for this, as all I have read so far directly related is the wiki talk page on instant messaging challenges. Whatever the current state and plans, it is clear that comprehensive solutions will be both necessary and feasible for large scale deployment. That has obvious implications for interconnection to standard email, SMS (widely used by the same age group with cell phones in developed countries, but much more convenient with child sized laptop keyboards and screens than with mobile phone keypads and tiny text windows), voicemail and fax (all of which are already  gatewayed to each other and to freely available social networking web sites). That in turn has obvious implications for non-educational uses of the same hardware. Some work on async messaging like email being made to appear just part of continuous networking at 21.

8. Another fairly obvious point is that with VOIP, each laptop potentially is a cellphone, when mesh connectivity avaiilable.

9. Less obviously, each laptop is a cellular base station that can connect ordinary cellphones to the PSTN in areas where there is no deployment of either CDMA450 or 3G base stations for mobile service. This potentially means that a village with an OLPC mesh could potentially also have cellphone connectivity earlier than would otherwise be possible.

10. That is because the next generation of cell phones enables use of wifi through ordinary wifi access points. Already fully standardized and initially available premium instead of standard models. The primary purpose of this is for mobilcos to displace fixed landline providers by routing calls over home broadband connections transparently using VOIP. The necessary QOS features for low latency telephony are included with the next generation of 802.11n home wifi routers that provide internet, and hifi audio/video distribution within homes. A secondary feature is that public access points and mesh services can also provide higher bandwidth connectivity than mobile base stations, where available (and are becoming ubiquitous in developed countries).

11. There are severe additional complications in achieving telephony levels of latency when packets travel multiple hops through an mesh network. Nevertheless the relevant QOS protocol features at the MAC layer of Wifi are actually standardized for Wifi phones independent of the MIMO stuff included in 802.11n and there are existing wifi deployments, for example at hospitals that achieve successful fast handovers between access point coverage areas. The Mesh Deterministic Allocation MDA probably won't be included with initial 802.11 but should be added and llikely to improve things greatly. There may be scope for layer 3 work within the WiFi SOC that could be important in handling (and rationing of charging for) different flows across mesh, regional and internal links.

12. Likewise there may be additional delays in using multiple packet network hops from a village to the nearest suitable Point Of Presence instead of lower latency physical layer channels.

13. Nevertheless it is already clear that OLPC deployments with adequate internet connectivity can be extended to villages at least 50km to 100km from Points Of Presence that already have a Satellite terminal using techniques already deployed such as Wifi Long Distance (WiLD) from Tiers.

14. Tiers already has several deployments in India, Ghana, Guinea Bissau and Philippines.

15. They have a USD $50 power controller that can be used to combine a standard solar panel, a standard battery, and standard wifi routerboard. They supply kits with parabolic antennas to be mounted on towers for fixed long distance links in remote areas with no local power.

16. This means OLPC is deployable to villages well beyond the range of a single CDMA450 celullar base station colocated with a satellite earth station POP. An immediate consequence is that store and forward messaging services, including fax and voicemail as well as email can be made available to places that cannot yet be provisioned with standard PSTN services due to the costs of deploying satellite POPs and cellular base stations.

17. Less obviously, OPLC hardware is deployable, without the benefits of IP connectivity to the global internet, but with full messaging capability, even where WiLD type links are not yet feasible. The motoman project already connects a number of schools in Cambodia via motorbike wifi. Even where roads are unavailable, messaging (including offline web access like woffle) can be provided via flash drives, as is already being done for several South African schools (though via road transport) by the Wizzy Courier project. The bandwidth of a daily 1GB flash drive is about 100kbs (but of course with 24 hour latency delays)! Daily DVDrom greatly exceeds broadband capacity. Disk media can easily be exchanged by foot traffic and vastly exceed any plausible bandwidth (as opposed to latency) requirements.

19. Low cost cell phone banking systems are already emerging, for example in South Africa. Currently only 22% of the population of Africa has postal delivery to the home, 60% can collect mail from a local post office and the remaining 18% have no postal service at all. The implications of being able to extend messaging services and banking beyond what is currently feasible for extending PSTN are quite major. Interconnection with existing networks is essential due to "network effect" (eg businesses in already connected towns only doing business via fax and expecting a phone number they can at least leave a voice message at as well as send a fax to). Fortunately that is relatively trivial once the accounting issues are understood.

20. There are a lot of subtleties involved in the appropriate software infrastructure to enable administrator free messaging by children in a "sometimes (sporadically) connected" environment. Things have to sort of "work" from the viewpoint of the user whether you happen to have any particular real-time connection available at any particular moment or not (and regardless of interruptions), without manual "clean up" of problems. This turns out to require the maintenance of complex replicated structures across various groups of nodes. Fortunately a lot of work has been done on that in projects such as Coda, Odyssey and Aura. Also Intermezzo. Well worth carefully studying all four and any related. Some of the concepts are closely related to the idea of "spontaneous branches" in distributed version control systems like darcs. Essentially a "sometimes connected" messaging system requires a multi-master directory replication system which has inherent, though solvable problems that will bite in a "no administrator" environment.

21. Some highly relevant R&D is being done at University of Toronto, Networked Systems Group and connections are being established, though only recently.

22. Caches important with maximum trickery - eg of varying levels of image resolution so high res from web page only got when needed but then supplied instead of low res when others later request the page). Work on such projects at 21. They would be more useful at metro ends of international and regional links than at school servers. Other more complex stuff should also be done at the far end - especially to store low priority async during peak periods and forward it offpeak and to extract usable text fron web servers. Reference various Indian papers and other challenged networks stuff.

23. Many other significant cost and performance for communications would be feasible if they have their own ISP oriented that way to deal with the various situatins and tariff structures. Would benefit from differential charging and/or quota enforcement for connection oriented reservations and priorities etc that are built into application APIs and understood by developers. Resource allocation mechanism could also be relevant to other constrained resources, memory, flash space, flash writes, cpu, mesh bandwidth, and especially battery and battery charging. Internal ISP also important because real issues will arise in teleco liaison with increasing density and usage. WiFi mesh is neat where nobody else doing it, but backhaul and larger scale penetration of the "last mile" both need proper engineering.

24. Actual resource pricing and mechanisms for granting allocations, paying and receiving could also work with other community currencies eg for mentoring/tutoring local work projects etc and with real currency and be part of education program and banking networks/ecommerce as in cellphone banking and indian kiosks.

25. SES participation probably indicates already interested in "satellite dish in every village" implications. Nortel also participating. News Ltd perhaps also thinking that way.

26. The journal stuff archived to google disk farms also has various implications.

27. Lots of ideas for things one could do with the audio and camera and touchpad etc. eg group of kids doing a survey with or without interviews and photos and mapping an area or a building.

28. Image expertize might mean they could tackle processing poor VGA camera images into usable scan of document for fax and OCR. Also more photographic pictures from multiple frames. Also precision 3 dimensional mapping from photos from different positions with scaling by including a standard length stick or triangle. Automatic feature recognition to correlate recognizable points, supplemented manually. Perhaps google does this sort of stuff with aerial photos. Also for face-id kits educational activity.

29. Forms already in XMPP protocols and detailed standard for administering, scoring, etc educational assessments. Could be applied to people surveys with rotation of alternative answers, question format etc, voice playback of questions and recording of answers etc etc.

30. Would like to scratch an itch by doing multi-master replicaitn protocol as Sugar activity over mesh with educational display as well as practical use.

31. Also forgetful bloom filters.

32.The python 2.5 library for processing XML as combination dictionary lists should be applicable to ASN.1 codec for improved presentation level efficiency.

33. Also such structures should cover generality of Typed Oriented Directional Attributed Hypergraphs.