Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory have created a system for testing new network traffic management protocols that requires no alteration to network hardware but still works at realistic speeds—20 times as fast as networks of software-controlled routers.
Existing testing protocols have hampered development and adoption of alternatives to the transmission control protocol (TCP). Data center routers are hardwired to run TCP. Testing requires either using reconfigurable chips—whose programming is labor-intensive—or software-controlled routers, which are too slow.
The new testing system, dubbed Flexplane, relies on a computational network model running the protocol to be tested. The system periodically schedules transmissions of test packets to the live network.
The TCP protocol can, when multiple data packets are queued up waiting for transmission, drop packets at the end of the queue. An improved protocol could use information in a packet’s data header to alert users to traffic jams and to assign priorities to packet transmission.
Flexplane models a network that uses a new protocol that transmits only packets’ header data and uses only the data relevant to protocol implementation.
In test mode, a server on the real network requests Flexplane for permission to transmit a data packet over a virtual network running the protocol under test. The emulator sends a dummy packet through the virtual network.
If the packet is successfully delivered, the emulator gives the live network permission to send a packet. If the packet does not reach its destination on the virtual network, the live network does not send a packet. Packets labeled high priority are bumped ahead of others.
“Being able to try real workloads is critical for testing the practical impact of a network design and to diagnose problems for these designs,” says Minlan Yu, an associate professor of computer science at Yale University. “This is a smart idea that achieves both high link speed and programmability.