Misc links

• XMPP news: http://www.jabber.org/pipermail/council/2004-April/001252.html
• Switching to Mac: http://polyphase.ca/switch/
• Repository Planning: http://wiki.osafoundation.org/twiki/bin/view/Jungle/RepositoryPlanningZeroPointFour
• Reid's new stuff: http://sourceforge.net/projects/wix
• Another PIM client and related server: http://www.kolab.org/

More existing repository access protocols -- focus on replication

How to choose a replication protocol: http://www.cs.mcgill.ca/~kemme/papers/srds01.html. This is a great paper with a counterintuitive result: the most obvious way to do replication for cluster computing may be the best way we know. The "most obvious way" (because it's probably been the most implemented is called ROWA, for Read-One-Write-All. That is, when reading replicated data, it's OK to just read it locally. When writing it, be sure to change all the data locations. Writes can be propagated immediately or deferred until the transaction is completed. This is compared to the Quorum approach where one replication site needs to get a quorum before being allowed to modify any data. The quorum model variants called majority, tree and grid are all compared.

The HTTP Distribution and Replication protocol - http://www.w3.org/TR/NOTE-drp-19970825. This protocol was designed by Marimba particularly for clients to download software from the Web. It extends HTTP by adding headers with IDs to make replication more certain, by adding index files to list all the files that need replication, and by allowing partial file download. There's some claim that it supports "push". It's by no means a standard even though it's published on the W3C site. It's rather disturbing that no mention is made of ETags, as it looks like this proposal overlaps heavily with ETags. It's rather underspecified.

HARP -- http://www.cs.cornell.edu/Courses/cs614/2003SP/papers/LGG91.pdf: Highly Availabile, Reliable Persistent file system. Uses the primary copy replication technique. Server oriented, uses replay logs to make changes.

Alternatively, HARP can be Hierarchical Asynchronous Replication Protocol but this seems to be dated research (from the 90s).

State Replication Protocol -- http://wiki.cs.uiuc.edu/CampSmalltalk/About+State+Replication+Protocol+(SRP) . SRP is pretty ambitious and has a lot of similarities to what Chandler is trying to do.

A peer-to-peer distributed file system: http://motefs.cs.umd.edu/. This doesn't seem to be making progress beyond listing goals currently and is research rather than a standard.

SyncML -- http://xml.coverpages.org/SyncML-WhitePaper.pdf. SyncML is server-based, and designed by the Open Mobile Alliance, which is generally a bad sign.

-- LisaDusseault - 05 Apr 2004