Dale Ghent wrote: > On Dec 1, 2008, at 4:24 AM, Rainer Toebbicke wrote: > >> With unlimited development resources AFS would deserve a better suited >> protocol than TCP, in practice with a little more realism my gut >> feeling is that at least some more brain should be devoted to >> improving plain RX rather than betting on another horse. I >> occasionally tried over the past years, with some improvements that >> Hartmut tested as well, but my brain being what it is and the matter >> relatively complicated results remain modest. > > Given this, what are your thoughts on STCP and whether it would be > useful in this arena?
>From my perspective there is still performance improvements that can and will be obtained from Rx UDP. There is active work being performed by Your File System Inc. in this area. Some of the results have already been pulled into current OpenAFS releases. Others efforts are still in a research and development phase. However, no matter how fast Rx UDP can become, it will not be able to overcome the problems caused by Network Address Translators and Firewalls that close outbound udp ports after 60 seconds. To address these issues we will require either TCP or SCTP. As Derrick points out, SCTP is not available on all of the platforms we support or want to be able to support in the near future. As a result some form of TCP based protocol is going to be necessary. Even if the protocol is nothing more than a tunnel over which existing Rx messages are transported. One of the biggest challenges we face is restructuring the use of Rx by the cache managers and the services to permit better utilization of the network. As an RPC based protocol, most operations are performed in a serialized manner. That is not a weakness of Rx but a weakness of the application model. Jeffrey Altman
smime.p7s
Description: S/MIME Cryptographic Signature
