What will OpenFlow do for Networking in 2013?
We are a little over a week into the testing phase of our State of OpenFlow 2012 report. One trend we are following is the entrance of new vendors focusing on hardware and limited software, relying on Network APIs for control.
On Friday we received a new OpenFlow based switch with 10G and 40G ports. We were able to confirm that the switch worked with Floodlight and RouteFlow in the same setup as the current switches we are testing. This shows a important moment in the state of the industry. With Network APIs and conduits such as OpenFlow, new entrants to the Router and Switch market can focus on delivering better hardware for less money.
More proof is seen when configuring the Pica8 Pronto 3290 for optimal OpenFlow performance, all of the configuration is done from a shell. There is less overhead as the CLI, Protocols and other features are not loaded. We believe OS features are becoming less important as the features become part of the OpenFlow specifications. The forwarding hardware is getting better and better as companies can spend less R&D on the network stack.
A New Vendor Asks to be Included in our Testing
We were contacted on December 19th by a network equipment vendor informing us that they had a switch they believed would be a good fit for our end of the year report. We picked up the switch on the 21st and started testing. Connecting the switch to Floodlight and RouteFlow was straight forward and within about 30 minutes we were running traffic across the switch.
Running line-rate traffic across switches
Something we at Router Analysis have been looking forward to is running traffic across the switches. Over the last week we have been running minimal traffic across the switches to confirm functionality. This week we started running real traffic across the switches. Our traffic is based on sampled routes and flows from Internet routers, combined with our own tools for flow aggregation and traffic generation.
Due to the design of the HP 6200yl we had to use switch ports for the management interface and uplink into the controller. As the HP only accepts HP SFPs of which we have four, we only have four interfaces to work with; leaving only two interfaces for traffic. We were able to run line-rate traffic across these ports.
As the Pronto 3290 has separate management and uplink ports and accepts any vendors SFPs, we were able to take advantage of as many ports as we could connect. We eventually stopped at 1000 flows and 40G of traffic. While we would have loved to push further, we believe this level of traffic shows that OpenFlow is a solid, functional way to control a hardware switch and meets our guidelines for the report. The report is about the state of OpenFlow, not a vendor comparison.
The Pronto 3290 is not the latest, or fastest switch Pica8 offers, but it is a workhorse. When we are pushing a lot of flows to the switch, the switch complains about dropping log lines and having its CPU consumed. This is expected. The switch still installs the flows we give it and pushes the traffic without issue.
As we get into the last week of testing for the State of OpenFlow 2012 test, the testing is getting more interesting, we are pushing the limitations of the hardware we have and having a lot of fun doing it.
New entrants into the market don’t have to spend the same amount of time developing their software interfaces to the hardware features. This lowers the cost of entrance into the market and we believe will drastically change the landscape in the coming years.