14:35]]ForCES Minutes, IETF 57th, Monday 07/14/03
Attendees: about 50
WG status:
The Requirements and the Framework documents should become RFCs soon. They
are currently under AD/IESG review.
AGENDA:
5 min - WG status - chairs
15 min - ForCES Framework respond to IESG feedback - Ram Gopal
http://www.ietf.org/internet-drafts/draf
t-ietf-forces-framework-06.txt
20 min - ForwArding and Control ElemenT protocol (FACT) - Hormuzd
Khosravi
Individual contribution protocol proposal.
http://www.ietf.org/internet-drafts/draf
t-gopal-forces-fact-04.txt
20 min Netlink2 as ForCES Protocol - Robert Haas
Individual contribution protocol proposal.
http://www.ietf.org/internet-drafts/draf
t-jhsrha-forces-netlink2-01.txt
15 min - ForCES Forwarding Element Functional Model - Joel Halpern
Individual contribution model proposal.
http://www.ietf.org/internet-drafts/draf
t-yang-forces-model-02.txt
*** NOTE: Authors intend to ask to make the above a WG
document. ***
*** Please read and review prior to meeting or comment on list!
***
15 min - XML/TIPC based ForCES protocol and ForCES
Pre-association Phase Protocol - Jon Maloy
http://www.ietf.org/internet-drafts/draf
t-nait-forces-xml-model-00.txt
http://www.ietf.org/internet-drafts/draf
t-nait-forces-pap-00.txt
15 min - ForCES FE Functional Model Elements - Zsolt Haraszti
Individual contribution model proposal.
Late contribution - available at:
http://www.petri-meat.com/slblake/networ
king/drafts/draft-haraszti-force s-model-00.txt
15 min - Demonstration of working prototype of XML/TIPC for ForCES - Jon
Maloy
Drafts:
Minutes:
Reached consensus that protocol selection will be attempted on the
mailing list in a month.
Comments on selection process, Joel Halpern spoke to first
determining if the protocols are doing what we want. He thought it is
obvious that they meet the requirements. Jason Goldschmidt asked if only
post-association will be considered. Answer was yes. Comment was made that
protocols should ONLY focus on post-association. Comment was made that
documents should be sure to use a consistent language, set by the
framework and requirements.
Framework Presentation (Ram Gopal)
Update 05 and 06 made changes to address FE virtualization. FE
partitioning is done during the pre-association phase. Comments were made
about multiple FE managers, these comments were not accepted. Next draft may
address offloading concern raised by IESG. Text added to support
graceful restart, non-stop forwarding and dynamic association
establishment. New section added to better detail of FE
loss/restablishment, caching and session reestablishment.
The new draft adds text to discuss security threat; see security
considerations of draft for more info. Security options for FORCES are as
follows: No security, TLS, IPSec. New draft incorporates security
handshake during association establishment.
No comments/questions on framework.
FACT (Hormuzd Khosravi)
Protocol updates. Now compliant with requirements draft 9.
New draft adds definitions for separate control and data channels.
Protocol Elements (PE) will be sent over data channel, all other FACT
messages over the control channel. This was done to prevent against DoS
attacks that would saturate the channel between FE and CE. Same
reliable transport used for both data and control channel. This
requirement is driven by the need for congestion control; though this is not
called out in the FORCES requirements document.
FACT uses reliable transport to meet requirements. TCP/SCTP is
recommended. This simplifies the protocol design.
Security rewritten. 3 modes, no sec., IPSec, TLS. TLS is
recommended.
Comments:
Joel: we need a mandatory to implement core of protocols for transport and
security. David Putzolu: pick one and go with it.
Description of CE failover and difference between weak and strong
consistency. Strong, FE communicates with both. In weak, CE's sync.
Question/comments
Zsolt Haraszti: problem with separation between data and control
channel. For example asynch exception will happen over control. What if
asynch exception requires sending the packet to the CE?
Joel: the kind of events that go over the control channel is regarded
about the FE itself. The data channel handles info about packet related
events.
Robert Haas: is the priority bits not enough to distinguish between the
two.
Joel: Data channel should not be reliable.
Xiaoming Fu, How does CE-CE work.
Hormuzd: This is out of scope given the FORCES charter.
Ram Gopal: FACT can be used for CE-CE.
David: Framework points out the channel and explains why these other
connections are out of scope.
Xiaoming Fu: Maybe next step in signaling WG and FORCES can work
together.
Netlink 2 Robert Haas.
Summary of changes.
Joel: is it your assumption that FORCES is used for FE-FE or LFB-LFB
communication. Comments to clarifying that netlink2 is not for LFB-LFB
communication.
Netlink2 in a nutshell.
Uses groups and multicast to allow for scaling. The Goal is flexible CE-FE
addressing. Netlink2 is making use of groups. Then some addressing
examples were given.
Conclusion. Netlink2 addresses scalability whereas FACT focuses more on CE
failover and availability.
Joel: scalability addressed by MC, this concerns him. IS there a need to
send the same messages to the different FE's? How common is this today
since FE's are not the same. Ie different routes, ports, yadda yadda. It
doesn't seem obvious that there is a value add in sending same
information to all or a set of FE's.
Patrick: if multiple FE of the same box they must have the same info for
certain things a good example is the forwarding info generated by the
routing protocols.
FE Model - Joel Halpern
State of document. Most of sections are there. Not all filled out.
Fine grained FE blocks. Does not want to talk about large scale blocks,
like a DiffServ block. How to represent an IP Forwarding block (this is why
there needs to be an affinity between NPF and FORCES).
FE block description. Formal description of a block is needed.
Classifier.
How to build a classifier? Two classifiers, one that modifies the
meta-data, other that has multiple exits. Compromise.
Meta-data. Where and how to use meta-data? FORCES model that can do
useful things, before meta-data definition is answered. Group need to
resolve how to tackle this.
Representation. Is representation for documents same as an on the wire.
Team recommends XML Schema. Not DTDs
Thinks document is consistent with the framework and possibly group.
Should document be made a WG doc?
Comment. Classifiers without metadata seems like science fiction.
Joel: having a forwarded that does such is more uncommon.
Jason: why not classifier plus mux?
Joel reiterates the need for meta-data definition.
David: Should this document be accepted as WG document? Hums? The Favors
have it. Still need to check with mailing list.
Pre-association protocol - Jon Maloy
FE/CE manager accepts/rejects new PE's.
Why do we need this?
Need for dynamically changes NE's.
Protocol discussion.
Robert: Could we use SLP instead of discovery phase proposed for CE-CE
Manager protocol?
Demo
Joel: we should focus on our chartered work, not out of scope work.
FE Model elements (Zsolt Haraszti)
not alternative model.
Topo versus encoded.
a mixture is needed to deal with the problems related to each.
|