ForCES Minutes - IETF59, Seoul, South Korea, Forwarding and Control Element Separation WG (ForCES) Tuesday, March 2 at 0900-1130 ============================= CHAIRS: David Putzolu Patrick Droz attendees: 62 (signed the blue sheet) AGENDA: 5 min - WG status & Agenda Bash - chairs 40 min - Transparent Inter Process Communication protocol proposal - Jon Maloy http://www.ietf.org/internet-drafts/draft-maloy-tipc-00.txt 20 min - A Testbed for GRMP Protocol - Dr. Ligang Dong 20 min - Updated Forces FE model (Steven Blake) All presentations from the meeting are available at the following web site: http://www.sstanamera.com/~forces/Ietf59/index.html 1) WG status Status: 1) Charter We need to update the charter. Please look at the charter and see if the dates are reasonable? Comments should be sent to the mailing list or David or Patrick as chairs. 2) Document status: a) Forces framework has been approved for publication as an RFC. b) Forces Model was just released in February - It's over 100 pages. - Should we consider splitting into 3 different drafts. c) Evaluation Drafts [draft-ietf-forces-evaluation-00.txt] - 3 protocols proposed, and an evaluation draft worked.- - Design team formed to determine if a merger is possible. Design team will report back on March 20th. - Please send comments to the list on what might be good. d) Applicability statement is on hold until further work completed on other drafts. draft-ietf-forces-applicability-02.txt 2) Transport Inter Process Communication TIPC is not a ForCES protocol but rather a transport protocol. It is freely available under Linux. It offers also reliable multicast. Discussion: Joel Q: asks 1) Whether the stack could support multiple channels (Ethernet, UDP, etc.) simultaneously, and if so how they would be selected. Jon A: it could support such, and that the decision was based on scope (a cluster used ethernet, inter cluster used other mechanisms.) Joel Q: asks 2) whether they had implemented inter cluster or internee support. Jon A: they had not completed that work. Joel Q: Is there a keep alive protocol assumed by TIPC? Jon A: Yes, there is an underlying protocol sending probes. There is a underlying timer that watches the probes to see if it comes back. David Putzolu Q: Did you determine a connection has a loss or is in error in the transportation? Jon Maloy A: There is a queue on each link. On this queue the buffers will be released once the messages arrives. when this queue exceeds a certain limit, then the messages will be tossed. The process trying to send the messages will queue messages if congestion is detected in the transport. On the receiving side, the messages will be rejected if the messages are in error. A messages will be sent back indicating the error. Hormuzd Q: Is it possible to address a message to multiple LFBs ? Jon A: This is possible for multicast, unicast will be on per LFB basis 3)20 min - A Testbed for GRMP Protocol - Dr. Ligang Dong Introduction to Testbed - deployment CE code using VC++ Smartbits software used to measure packets Introduction of the experiments on the testbed: 2 experiments - one for configuration of LFB, second for redirection packets/DoS attacks 2 scheduling schemes - FCFS, priority based, rate based with RR experiments with scheduling strategy from linux kernel Hormuzd Q: Did you experiment with different packet size? Dr. Dong A: answer; yes. David Putzolu Q: Did you try a congestion aware protocol for the control channel? Dr. Dong A: Yes, we did. The results (table 3) on the slides. The information is partially complete. We will be doing more. Table 1 - is about the loss of control packets. 3) Updated Forces FE model (Steven Blake) [20 minutes] http://www.petri-meat.com/slblake/networking/drafts/forces-model-02.pdf 3rd revision of document. David Putzolu Q: The url does not work for the XML Steve Blake A: Perhaps we can get this working. Joel Halpern Q: It would good to have a location to get all the XML for forces. Steve Blake A: May be the working group can provide a location to put all the XML on. We don't really know where to put XML. MIBs are contained in files. Joel Halpern: we are looking for feedback on the XML information. Open issues: 1) Is the augmentation useful 2) How useful are input groups? 3) in which documents will the protocol payload associated with the LFB attributes be defined? - forces model? - LFB Class definitions? - Protocol specifications? Pending work items: (be ready for last call in August) 1) Define rules and schema extensions for LFB class derivations via inheritance 2) Support for data type augmentation (if needed) 3) Inclusions of type and Class IDs (if needed) 4) Rules for version handling 5) Simplistic handling of LFB class definition using the LFB Library schema 6) Improve schema representation of capability attributes for an associated attribute. Discussions: Lily: 1st two items are related The primary reason we have data type augmentation is for inheritance. We'd like feedback on that