Connecting Mobile Workstations to the Internet Over a Digital Cellular Telephone Network

Markku Kojo, Kimmo Raatikainen, Timo Alanko, Univ. of Helsinki

Main points:

• TCP/M-IP compatible indirect-TCP implementation using "Mowgli sockets" API. Mowgli Socket Interface Protocol is like ICMP for wireless link; Mowgli Wireless Link Proto does stuff like header compression, etc.
• Mowgli Virtual Socket Layer demuxes stuff to either Berkeley sockets or Mowgli Wireless Msg Protocol/Mowgli Socket/MWLP stack. (See fig. 5 in paper.
• "Mobile-aware" apps can use additional mobility support in M-sockets, e.g. "QoS interface" (proposed but not defined).
• Claim: proxy is not a bottleneck; slow wireless link is (by defn).
• Future work: distinguishing different qualitative types of communication service needed by apps, and develping QoS models; specification of services needed for mobility awareness; security (no elaboration given).
• Great architecture for dealing w/disconnection, if you're willing to do that at the transport layer - an open argument.
• The Mowgli Socket Protocol they talk about seems to have evolved into the MWLP/MDCP described in the Mowgli-WWW paper summary.

Flaws/comments:

• QoS discussion lacking - no direct specification of the QoS params (but then, neither do we...)
• Indirect TCP was done before and has been shown to not necessarily be a good thing; main point here seems to be to provide Mowgli sockets as the "preferred" MH interface, but to allow for a compatible interface as well.
• Not clear if the transport layer is the place to deal w/disconnection; isn't that more an end-to-end problem that applications should solve? (Though GloMop deals with part of it in session layer, it notifies apps of disconnection, and can serve a disconnection request from an app.)