XS Server Services

From OLPC
Revision as of 21:54, 22 August 2007 by Wad (talk | contribs)
Jump to: navigation, search
  This page is monitored by the OLPC team.
  english | 한국어 HowTo [ID# 61073]  +/-  


These are services that the School server will provide. Additional services under consideration for deployment are listed separately. Service meta issues such as installation and management are discussed on School server specifics. A detailed description of the current implementation of these services is also available.

Please help by adding links to existing pages discussing these topics, if you are aware of them

Library

The school library provides media content for the students and teachers using a School server. This content may either be accessed directly from the school library or downloaded onto the laptops. The content in the school library comes from a variety of sources: OLPC, OEPC, the country, the regional school organization, other schools, and teachers and students in the school. It includes both software updates and repair manuals for the laptop.

A number of library attributes (self-management, scalability, performance, identity) are discussed in their own sections below. But it is clear that the library is not a single service, but rather a set of services (wiki or wiki+,http cache, bittorrent client, codecs, etc...) which combine to provide the Library functionality.

The library at its core is a "local copy" of a library assembled by the school's country and region, with help from OLPC. This set of documents is assembled on a country/region server, and is named a Regional Library. Each School server has a local library consisting of documents submitted by students and teachers using that server. This is named a Classroom Library. The collection of Classroom Libraries provided by the School servers at a particular school (there will typically be 2 to 15) is named a School Library.

Discovery

Basic bandwidth assumptions: there will be sporadic connections between XOs via the mesh, fast connections between XOs and school servers [SSs], possible sporadic connections b/t SSs and the internet at large via regional servers [RSs], and generally fast connections (at least fast downloads) b/t RSs and the internet.

Notion: XOs can maintain an index of materials seen on local XOs, school and regional servers; and can send requests for material over the mesh or network. School servers can maintain a larger similar index, collating requests for materials and evaluating whether they are likely to be fulfillable over the mesh. Those that aren't can be queued for request from regional servers or the internet.

Indices would include a variety of useful metadata, including size, location, latency/availability (available via Torrent, FTP availability, estimated speed of connection), and rough content-based IDs.
Servers and perhaps XOs would maintain basic statistics on what materials and types were most requested, with some special attention to requests for large files (which are harder to fulfill without saturating the network), and when nearby queues and indexes have been updated.

A reasonably accurate method is needed for keeping these indices mutually up to date, sending feedback back to requestors ("this material is 9GB in size; the request is unlikely to be fulfilled. please contact your network maintainer for more information."), and sending feedback to content hosts ("this file is very frequently requested. please seed a torrent network with it. <links>").

On the "internet at large" side of indexing are national and global libraries, metadata repositories such as the OCLC, global content repositories such as Internet Archive mirrors, and a constellation of web pages organized by various search-engine indexes [cf. open search, Google]. There is fast connection among these groups, and various categorization, discussion, and transformation of material (for instance, to compressed or low-bandwidth formtas) can be done there in preparation for sending it back to the less swiftly-connected parts of the network. (For instance, one often scans high-bitrate multimedia on a local networks at much lower bitrates, edits the low-bitrate material, and sends the collected edits back to be applied to the original media high-power machines and over a high-bandwidth network.)

Scalability

A pre-determined subset of the Regional Library is mirrored in the School Library, the remainder of the documents are available over the Internet connection.

Caching of recently accessed documents will be done to provide the best user experience given limited connectivity, but it also has the effect of minimizing the load on the regional library. This will be performed by the transparent HTTP proxy.

While a peer-to-peer protocol may be used to help scale the Regional library, it must be sensitive to the fact that a large number of schools will have very limited upstream bandwidth and would not participate as peers. And while this protocol may serve well for mirroring content onto the school servers, using it for student accesses to the library would require a more complex client than a browser (see P2P Caching).

At the school level, there is some possibility of taking advantage of local connectivity to distribute the document cache among all servers in the school. Unfortunately, the exact piece of software we need hasn't been developed yet (see P2P Caching).

Self-management

What is the mechanism for user submissions of media to the library? Teachers (and students) should certainly have a place to put resources they create where all can retrieve.

What about local wikis? Should they be supported on the School server? They are probably best replaced with a more function oriented information organization tool such Rhaptos or Moodle, and an online extension of the journal which supplants blogging.

Repository

Content centered, but allowing collaboration in building new content. See http://cnx.org for an example. The service is provided as open source software by the name Rhaptos. It's Python based.

Collaborative/Publishing/Learning Management System

Moodle could be an option? http://moodle.org

Exchanging published work with other communities could be done using a USB key, taking a class portfolio (using moodle pages) to another village. Distance collaboration can also be done in an asynchronous way, where each school has its page or set of pages in Moodle, and it would work like standard mail, travelling by land to another school, kids contributing, commenting, peer-assessing, and then back again.--Anonymous contributor

A Mini-Moodle Plan

These draft notes from Martin Langhoff, Moodle developer (14/03/2007): we are keen on helping prepare / package / customise Moodle for deployment in this environment, and have experience with Debian/RPM/portage packaging and auth/authz/sso. Actually, I lurk in laptop-dev and sugar-dev keen on discussing this further. Right now reading up on Bitfrost.

There's been some private correspondance between Moodle devs (MartinD, MartinL) discussing an overview of what the main work areas for Moodle on XS would be:

  • Packaging (RPM?)
  • Provide an out-of-the-box config that Just Works
  • Integration with auth / sso (Bitfrost)
  • Integration with group / course management (Is this too structured ;-) ?)
  • Replace mod/chat and messaging infrastructure with mod/chat-olpc that hooks into Sugar chat/IM
  • (maybe) Replace mailouts with OLPC infrastructure for email
  • Work on an XO theme and UI revamp

(MartinL is happy to help on all except the theme/UI, due to sheer lack of talent)

That's the basics of getting a basic integration in place. The fun work will be in creating activity modules that make sense in a XO environment.

Maybe this should be moved to a subpage?

Bandwidth Amplification of Local Content

This refers to the problem of supplying content posted to a School Library to a large number of other schools. This is a problem due to the limited bandwidth available for uplink from a typical school. Unfortunately, the architecture most useful for nonprofit content distribution --- peer-to-peer (P2P)--- is not well suited for use by school servers using highly asymmetrical (DSL, satellite) network connections.

One solution is a well-connected server with a large amount of storage provided at the regional or country level which will mirror the unique content from each school. This has the dual purpose of backing up said content. P2P access protocols may be used to reduce the load on these (seed) servers making use of any regional schools with good uplink connectivity.

Security and Identity

The Library uses the identity convention introduced by Bitfrost. When a laptop is activated, it is associated in some way (TBD) with a school server. This is the School server hosting the Classroom Library that a student is allowed to publish onto. Students not associated with a Classroom Library only have read permissions on the Library.

Backup

According to this description of the Journal, it will provide automatic backup to the School server, with a variety of restore options.

What are the plans for providing additional storage to users of the XO laptops? How does the Journal handle filling up the available storage on the XO? allocated storage on the School server?


Network Router

The School server is first and foremost a node in the wireless mesh which provides connectivity to the larger internet.

XS Usage Dual.png

A starting assumption is that it is largely a transparent router. It does not perform any network address translation, and very little packet filtering. It will perform bandwidth shaping to ensure fair access to the internet.

Providing IPv6 connectivity from the mesh to the internet will require the server to provide a tunnel over IPv4 to the closest IPv6 network, since most sites will not be providing direct IPv6 connectivity. --cscott

HTTP Caching

The only packet filtering proposed is a transparent proxy on port 80, which will allow a caching of commonly accessed HTTP (web) content to occur locally. This will reduce the load on the internet connection, as well as the response time seen by a user. This may be relied upon to implement the School library.

Again, due to IPv6/IPv4 interoperability issues, the proxy can't be transparent, unless we're willing to do NAT for the laptops to give them a routable IPv4 address. --cscott

The problem here is availability of HTTP caching software. The most popular open source provider, squid, is not IPv6 compliant yet. However, there's work underway for that.

A more intelligent caching/redirection scheme may be provided by CoDeeN. Unfortunately, it is built on top of commercial proxy software.

And a third option, (actually potentially related to CoDeeN), is Vivek Pai and Anirudh Badam at Princeton have done some work on a light weight caching proxy (HashCache) that may become suitable for OLPC at some point in the future.

See also Server Caching

Name Resolution and Service Discovery

The School server will use and support the use of Zeroconf techniques for device name declaration and service discovery.

It should also support traditional DNS by exporting the mDNS name registry.

mDNS has some scaling issues which do not (yet) seem to be resolved. In the IPv6 world, we'll try to use Router Advertisement and well-known IPv6 anycast addresses to provide services without the need for explicit discovery protocols. --cscott

Bandwidth Fairness

The School server should implement a bandwidth fairness algorithm which prevents a single user from dominating the use of the internet connection. The problem is that the short-term fairness algorithms used by TCP give P2P software equal consideration as users trying to access a document over the web. By biasing the queueing algorithm based on usage over the past 4 to 24 hours, we can protect the random browser from the heavy downloader.

The problem is that due to our mesh network, we may really be penalizing a remote classroom relayed through a single node.

We could bias the bandwidth allocation based on remote port (e.g. favoring web access to port 80), but this solution seems less than optimal.

Dynamic IP Address Assignment

The school server will take responsibility for assigning IP network addresses to device on subnets it is routing.

Does the school server give out IP addresses to non-OLPC hardware? If so, should the DHCP be linked to the mDNS name registry ? I.e. if a machine requests a particular name in the DHCP request, should this name be recorded in the mDNS registry for the subnet?

IPv6 address assignment is magic; we may consider using the SEND protocol to ensure that routers are trusted. We will probably be using some form of temporary address in IPv6 for privacy reasons, which will require DynDNS or some other mechanism to allow friends to find each other. To protect kids' privacy, the solution can't be a permanent DNS name for each laptop; we need to more flexibly create task/address mappings so that friends can find each other for an activity without making themselves world-locatable. --cscott
Things to think about: Mobile IPv6, IPv6 privacy extensions, SEND (secure neighbor discovery protocol). --cscott