We open this final post on the State of OpenFlow 2012 testing with a very specific statement: OpenFlow is positioned to succeed in 2013. The final report, which should be out early to mid January will contain specific information backing this statement past what we talk about here.
The State of OpenFlow in 2012
OpenFlow is here. With Pica8 coming out of stealth, HP, Juniper, Cisco, Brocade, IBM, Huawei and IWNetworks putting themselves out publicly supporting OpenFlow. The growth of BigSwitch Networks and their Open Source OpenFlow controller Floodlight. Quagga’s addition of a southbound API to feed FIB entries into other products such as RFServer (part of the RouteFlow project). We definitely see a lot of solid support for OpenFlow.
Out of all the products we tested, the most refined and functional product comes from a relatively unknown player, Pica8. Pica8 and their Pronto line of switches have been behind many of the public success stories about OpenFlow deployments. Pica8 is part of Google’s OpenFlow network. Pica8 is heavily involved in the academic and research space, working with the RouteFlow team, InCentre, the Ryu Open Source OpenFlow controller and many others. Pica8′s 3290 switch is possibly the most tested OpenFlow switch in the field today.
Router Analysis used the Pica8 Pronto 3290 as our reference platform when confirming any issues we had while testing.
The oldest product we used for testing was the HP 6200yl. Introduced over six years ago, the ProCurve 6200yl is a solid switch with a good CLI. HP has been actively updating the way that OpenFlow is configured on their switches which shows that they are listening to feedback and reacting. While we were limited by the number of ports we could use, the HP ProCurve 6200yl passed all of our tests without issue.
The newest product in our test was the IWNetworks SDN 8952S. The physical design of the switch is well done, with multiple management ports, a usb port for copying files and redundant power supplies. The switch supports OpenFlow without issue and was able to complete both our Floodlight and RouteFlow testing. While the switch is 10G/40G it had no issue accepting any of the optics we put in it, allowing us to utilize as many ports as possible.
Our test plan was not complex and was not meant to stress any of the boxes. We did push the Pronto 3290 switch a bit as we had excellent support from the team over at Pica8,other than some tips on how to set the box up to boot into what ever mode we wanted we didn’t need any major support from Pica8 as the switch just “worked”, holding over 4k flows and pushing 40Gbp/s of traffic. There are some caveats we need to point out: We had to prime the switch with traffic before loading it up. The Pronto 3290 is an older switch and it’s CPU does not appear to be that fast. We ran into issues where the 3290 would drop logs due to CPU consumption. We also ran into issues where we over-ran either the switch or the controller when attempting to start up a lot of flows at the same time. We plan to release some numbers in the coming weeks regarding TTFP/TTFB.
The HP 6200yl supported 1200 flows in Hardware and more in Software. This met our needs and allowed the box to perform well in our tests. We saw no issue pushing 2Gbp/s of traffic across the box. We saw no problems with CPU strain, but we also did not push the 6200yl very hard.
The IWNetworks SDN 8952S was the most powerful of the switches we tested. As it is a very recently announced product, we were pleased to be given the chance to put it into our State of OpenFlow 2012 test. OpenFlow support is solid and the switch showed no issues with performance during our tests. We did not have enough machines or port to really test the throughput of the SDN 8952S, but based on what we know about its architecture, there is no reason it cannot do line-rate.
We did work with other vendors who did not want to be identified at this time. When we finished the testing all boxes had passed both the Floodlight and RouteFlow tests.
It was a lot of work and a lot of time but Router Analysis, Inc. is happy to have had the chance to do a final round up of the State of OpenFlow at the end of 2012. We believe there will be a lot of disruption in 2013 and beyond. New vendors will enter the market using commodity chipsets and their own software. As orchestration is being handled by OpenFlow controllers such as Floodlight, NOX and RYU, the internal configuration of the switch is becoming less important.
We here at Router Analysis, Inc. want to thank all of the parties involved (and those who chose not to be involved) for taking the time to work with us and focus on the community. We are already scheduled months into 2013 with testing and excited at the information we will be able to bring to you in the future, through our website, SiliconAngle, Wikibon.org, Twitter and any other public channel we can utilize. We will also be contributing to major tech journals and discussing testing, Network Function Virtualization, Software-led Infrastructure and other topics that excite and drive us.