User login

Dimeji Fayomi's blog

06

Oct

2016

There isn't much to report, I worked some more on the RheaFlow paper to get it within the 6-page limit. I should have another draft out by this week.

28

Sep

2016

I spent last week writing a rough first-cut draft of the RheaFlow paper to be submitted to SOSR. I intend to finish the draft this week and send it out for reviews and comments. I will spend the rest of the week, revising the draft and sifting away unnecessary bits.

21

Sep

2016

I added code to provide interswitch link configuration for RheaFlow and performed some tests to confirm it is working fine. Based on the code, RheaFlow will now support an arbitrary number of switches all connected to each in a full mesh. I also played around with the Atrium SDN router to examine its features and make comparisons to RheaFlow for the paper to be submitted to the SOSR conference.

This week, I'll be working on a rough draft of the RheaFlow paper to be submitted to SOSR.

13

Sep

2016

Last week, I added a configuration to RheaFlow's configuration file to make TTL decrements on the switches optional because the some of the tests conducted showed that some OpenFlow switches are not able to perform this action in hardware. I also conducted more tests with the Pica and ZodiacFX switches, both switches are unable to process IPv6 related flow rules because in the case of the ZodiacFX switch IPv6 is not fully supported while the Pica requires multi-table rules to process IPv6 related flows because the IPv6 fields are too large to be processed by the switch hardware in a single table. I also tested multiple switch configuration with RheaFlow in slow path mode with two switches.

I'll be working on the code to handle packet forwarding to the controller via inter-switch link this week and verifying different configuration mode with multiple switches. I'll also get started with writing the paper for the RheaFlow.

07

Sep

2016

I updated RheaFlow's to code to make all compoenents start together using systemd. I also tested RheaFlow with Allied Telesis AT-x930 switch. While testing RheaFlow on the Allied Telesis switch, it was discovered that the switch does not perform TTL decrements of IP packets in hardware which impacts forwarding performance, I've started updating the RheaFlow code to include an optional configuration to order OpenFlow switches managed by RheaFlow to decrement TTLs.

I fixed RheaFlowProcessor.py's handle_packet_in method to log and discard packets received from the virtual switch's (dp0) OpenFlow LOCAL port. This was considered in the early stages because dp0's instances in my dev environment did not send packets from their OpenFlow LOCAL ports.

This week, I'll complete the updates for the optional configuration for TTL decrements and conduct further tests with the Pica and ZodiacFX OpenFlow switches. I'll also start work on completing the sections of the RheaFlow code that handles inter-switch link configuration and forwarding.

01

Sep

2016

Last week, I completed the refactoring of RheaFlow code. I also fixed some bugs introduced as a result of the code refactor. This week, I'll be writing a proper script for RheaFlow so that NetlinkProcessor component is started before the rest of the application. I'll also test it against available OpenFlow hardware switches.

22

Aug

2016

Last week, I modified RheaFlow's directory structure to be compliant with the directory layout required for installation via pip/setuptools. I also started refactoring the code to comply with PEP8 standards, remove debug codes and commented out codes.

I'll complete the code refactoring this week and test it to make sure the application's behaviour has not been changed because of the code refactorisation. I'll also start working on a launch script for RheaFlow to ensure all components are started together.

11

Apr

2016

I made changes to the RheaFastpath code to reduce the number of rules installed and I've started working on IPv6 route conversion to OpenFlow rules. So far, I've established connectivity between an IPv6 host on the OF switch and its mapped port on the Rhea VS(dp0). I've encountered a bug where IPv6 routes are not being dropped in the code and I'll be fixing that this week and getting started with mapping recieved routes to installed OpenFlow rules.

04

Apr

2016

I got fastpath working on Rhea last week and also modified my code such that interfaces added to the virtual switch are one end of a VETH pair created during the mapping process rather than OVS internal interfaces that are created and added. This allows packets hitting those interfaces on virtual switch(dp0) to be visible to the Linux kernel. I also had a discussion with Richard Sanger regarding the fastpath implementation on Rhea and he suggested that the rules being installed were too specific and would increase the number of rules installed on the switches, this situation may have unintended consequences as hardware switches have limited capacity to process OpenFlow rules.
I will be making changes to the fastpath code to fix then this week and proceeding to add the conversion of IPv6 routes to rules.

31

Mar

2016

I don't have much to report this week as I haven't been able to get fastpath working properly. I've fastpath labels properly applied and I've been able to install rules for the fastpath ports but the issue I have is that ARP packets sent to the controller by all switches are not delivered properly. Debugging continues...