From OpenFlow Wiki

Contents 1 Meeting Minutes for SWAI, 08 October 2009 1.1 Present 1.2 Actions from last meeting 1.2.1 Scoping dpctl changes for slicing 1.2.2 Slicing updates 1.2.3 Testing Framework 1.3 Other 1.0 Release Discussion 1.4 Other matters 1.5 Actions

Meeting Minutes for SWAI, 08 October 2009

Present

Bob L, David E, Guido, Yiannis, Tatsuya, Dan T

Actions from last meeting

Scoping dpctl changes for slicing

• Dan indicated the scoping was mostly done; waiting on the final openflow.h changes so as to build on those.
• Don't expect this to be the long pole in the schedule for this.
• Update since meeting: Plan to have the initial pass done by Tuesday 10/13.

Slicing updates

A long discussion that touched on the following points:

• Rob S is concerned about rate limiting to the CPU.
  • What is the OF abstraction of the port interface to the local CPU?
  • Would like queue queries of that port to respond accurately about queuing capabilities there if present
  • The focus should probably be on the datapath to control path interface, rather than the switch to controller as the first is almost always likely to be the bottleneck
  • We think some vendors are adding extensions to support this, but they may be making sequencing assumptions about the actions
  • Bob L pointed out that the CPU should probably take some responsibility for taking care of itself rather than trying to put the burden on OpenFlow.
• HP will not support the enqueue action; will provide a mechanism for respecting Priority/TOS bits.
• NEC will support both enqueue action and Prio/TOS
• Yiannis pointed out: Since the commonality is respecting Prio/TOS, will that move people in the direction of using that?
• Testing of slicing
  • Have reserved 3 machines (nfcm32-34). Yiannis went through this. One generates over subscribing traffic, one is the DUT, one sorts the output to verify that the bandwidth configuration was respected.
  • Guido asked: How can we deploy this in the near term in our network for validation? Expectation is that we will use flowvisor to proxy the enqueue actions.
  • Goal is to have Brcm implementation running reasonably soon (hopefully soon enough to be used in validation deployment prior to release).
  • Need to make sure someone takes the responsibility of verifying that the feature is working in the test deployment.

Testing Framework

• Tatsuya is looking at the Python simple controller project; it's running in 0.8.9 right now.
• Figuring out how to incorporate that or pieces of it into the test framework
• Currently vconn and rconn are used there; it was recommended to ditch that and use

straight TCP sockets

• Simple socket code from Glen reviewed by Tatsuya
• We discussed the scope of where the packet processing code should lie and how general it should be.

Other 1.0 Release Discussion

• We discussed the development and integration plan
  • Get wire format compatibility as baseline (with a controller)
  • Develop different features on independent branches
  • Ideally, integrate branches serially and test each as we go; more likely some will be grouped in the interests of time.
• Release criteria for feature testing is running on the local network with lots of general users and Glen hammering with some experiments.

Other matters

• Call for interest in further discussing 1.0 development and integration with Glen in the near future. Talk to Guido
• As above for talking about testing with Tatsuya, Friday at 2:30.

Actions

• Expect update on slicing progress (should be close to running initial tests by the next meeting).
• Update on additional discussions about testing.