From OpenFlow Wiki

Contents 1 Meeting Minutes for SWAI, 01 October 2009 1.1 Present 1.2 Actions from last meeting 1.2.1 Testing Framework 1.2.2 Raw Sockets 1.2.3 C Code Conversion Issue for Test Framework 1.3 Slicing 1.3.1 Configuration 1.3.2 Implementation 1.3.3 Testing 1.3.4 Delivery 1.4 OpenFlow Packet Taxonomy 1.5 OF 1.0 Updates 1.6 Actions

Meeting Minutes for SWAI, 01 October 2009

Present

Glen, Mikio, David E, Guido, Yiannis, Tatsuya, Dan T

Actions from last meeting

Testing Framework

• Tatsuya gave brief update on progress;
• Another week for development before a more complete update
• Discussed mechanism for generating OF messages; will examine Python simple controller

(in git:openflow-projects/simple_controller

Raw Sockets

Glen sent out the Python code demonstrating raw socket use. It came with a few grains of salt (use at own risk). We discussed a bit about libpcap and reviewed the decision about not to require it; kept with that decision.

C Code Conversion Issue for Test Framework

• Plan: Use swig against openflow.h (and other headers if necessary).
• Revisit in the future if it looks like this will be a problem
• There's an example of the use of this from the Python simple controller mentioned above

Slicing

With the configuration protocol now being optional, the queue configuration needs to be revisited. We also discussed aspects of the spec details and then dove into questions about the implementation.

Configuration

• The key information needed by the controller is: What is the current queue configuration? What are valid queue numbers for which I can issue enqueue actions.
• The information about the discipline (min BW) and current discipline value (% of BW) may not be exposed to the controller, but it will probably be expected.
• Add/modify/delete operations will be done outside the scope of the OF spec. Thus, the controller will not be able to set the min BW value of the queues from within OF.

Implementation

• Be clear in spec about VLAN prio or TOS bits and relation to queue operations
  • Probably a capabilities bit that indicates can map to queue without changing VLAN prio or TOS
  • Another "capabilities" bit (it's a feature, not a bug) that indicates that the enqueue action will result in a change to some packet field such as the VLAN prio or TOS bits.
• NetFPGA will not support queuing in the near term; this is why we need to implement queuing on the software switch
• Traffic control, tc, will be used to provide the interface to queuing in the linux kernel. Yiannis gave an overview of this. Will use the classless FIFO discipline and the classful HTB (hierarchical token bucket) disciplines. The priority bits in the SKB will be obeyed by tc (it's believed). So those will be used to convey the queue.

Testing

We discussed the issue of testing. There will be some instrumentation required to really test this. Goal is to have black box tests that verify this feature. Would almost certainly require some wiring configuration and cleverness with the current hardware.

Need to clarify validation requirement prior to initial release as full tests probably won't be available prior to then.

Delivery

Yiannis said the initial goal is to deliver the implementation by November 10. Guido pointed out that code complete is supposed to be done by November 3. We'll say the difference is validation.

OpenFlow Packet Taxonomy

We didn't have time to discuss Mikio's detailed packet description.

OF 1.0 Updates

Other than the slicing updates abovbe, we didn't have time to discuss additional OF 1.0 update status.

Actions

• Dan to spec the dpctl changes needed for slicing configuration; provide deadline
• Tatsuya to provide update on test framework development
• Yiannis to provide slicing spec update