On Wed, Mar 08, 2006 at 04:16:10PM +0200, marcelo bagnulo braun wrote: > > Others?? > In addition to the interface between the shim and application, it is imperative that the interface between shim and transport protocol be considered. At the time when the shim decides to failover, the transport protocol should be kicked so that it can reinitialize the state associated with estimation of various path properties (congestion window, RTT, path MTU, etc). Research has indicated that not doing this can be problematic both for the end host and the network [1]. This has been discussed in TSVWG and TCPM, and is described in at least a couple of internet drafts (in the process of merging): https://datatracker.ietf.org/public/idindex.cgi?command=id_detail&id=12091 https://datatracker.ietf.org/public/idindex.cgi?command=id_detail&id=10165 -Wes [1] http://gltrs.grc.nasa.gov/cgi-bin/GLTRS/browse.pl?2005/CR-2005-213838.html -- Wesley M. Eddy Verizon Federal Network Systems
Attachment:
pgpm0WuBlKtaj.pgp
Description: PGP signature