IETF 64 ForCES Meeting in Vancouver
Agenda bash
-David Putzolu (chair)
wg status
Protocol draft WG LC
Joel has requested for IESG review of draft, need to finalize other drafts before IETF LC
Base LFB library
-Joel Halpern
connectivity lfbs
packet validation & manipulation lfbs
classifier lfbs
packet control lfbs
queue & scheduler lfbs
David P - reuse from MIBs, PIBs, NPF, etc.
Joel - sure we can reuse, there is lots of related work out there
Weiming - how do you communicate between FEs, one being queue & other being scheduler
joel - model doesnt define how to communicate but who communicates, the actual communication is implementation specificjoel - control info is represented using metadata
ForCES MIB
-Robert Haas
Management of entire NE
Not duplicate information available via ForCES
SNMP may or may not see multiple CEs
Single NE MIB alternative & CE MIBs alternative
Furquan - do we consider the case where there will be more than 1 active CE in the NE?
Robert - yes, provided the functionality is not replicated between CEs
Jamal - why do we prefer the multiple CE alternative?
Robert - how do we implement the other option, need to define new protocols to support this
Jamal - how about using AgentX for multiple CE alternative?
MIB Contents
Avri - which FEs are being referred to? how about FEs that are down?
Joel - cant expect the MIB to tell you what FEs it doesnt know about, i.e. has been associated sometime in the past
Robert - agrees to add this to the draft
Avri - should we make this a WG draft?
Jamal - MIB should reflect the FE protocol object
Joel - MIB is controlling the CE, so it wont work this way
ForCES TML Service Primitives
- weiming
Joel - why do u differenciate between attributes and capabilities ??
weiming - agrees
motivation for standardizing SP
SP - APIs...
Joel - dont need to define the parameters, that is the implementation
Jamal - you need to define how its done
Joel - what are the services that TML needs to provide, not apis
Jamal - can we use socket primitives?
Joel - not comfortable defining apis, socket calls in the ietf
weiming - may be use different names?
Joel - doesnt agree this (TML Config) is the primitive
Joel - should not define subroutine for these functions
TCP+UDP TML
-weiming
motivation
TCP+UDP - good deployment
No TML messages
UDP for redirect & HB msgs
Joel - strongly recommed not to invent a CC or DoS prevention mechanism
?? - something is possible if we can look at the TCP queue size, but this is research
univ of amsterdam (prof??)- need to careful about this before sending it out on a real network
Joel - udp tml should only be used when there is no congestion in the environment, should be indicated in the applicability statement as optional tml
Jamal - single hop environment such as chassis may be suitable for this tml
Joel - how will multicast work for udp tml
weiming - have pl configure the tml multicast channels
jamal - can use multicast within one hop environment
joel - what is a master ce?
Intra-NE routing mechanism
- x. Guo
motivation
joel - terminology change, have multiple forwarding tables not routing tables
guo - agrees
jamal - should be merged with the existing topology draft, similarity, end goal is the same
david - why are security requirements different?
hitachi?? - need packet format for transmit packet from one fe to another
huawei?? - 2 cases, single hop & multihop?
jamal - one of the lfbs presented by joel is intended to define how packet traverses
joel - several different issues - need to provide a way to model different approaches, rather than standardizing the packet format
ForCES Implementation Experiences
- Jamal
Using ATCA based hw setup
LFBs to support IPSec processing NE
FEO
FEPO
redirect
netdevice/port
IPv4
IPSec
David - what TML have you used?
Jamal - TCP tml, will probably try udp for redirect shortly
ForTER - ForCES Router implementation
- Ligang Dong
Funding NSF china, 3 sources
More than 10 students working on this
Architecture very similar to NPF
Use Intel SDK for FE
Use ATCA chasis to build the ForCES NE
LFB model based on Intel SDK
|