If at first you don't ...

If at first you don't ...

If the first OLPC software release contains a robust but resource intensive update distribution scheme (ie, the rsync based proposal) that does not scale well as the deployment grows, continued development on the mesh based scheme ought to continue full speed, and sent out to the OLPCs as updates once it's developed and tested. The two methods would have to co-exist on the servers for a while, but the load from the older method would decrease even as the number of deployed OLPC units goes up.

Someone (in another comment to a comment) compared one of the proposals outlined at the end of the article to bit torrent. On the surface, it is similar, however an OLPC will send the update packet to all of its peers, not just the ones that ask for it, and peers receiving the update packet passively need not acknowlege or retransmit the packet (except in the sense required by the mesh). A single transmitted packet might reach many peers, as it's a multicast; bit-torrent uses unicast, and puts a higher load on the network bandwidth for multiple (>2) peers.