Abstract:
With network port speeds increasing all the time, and the need to process more data everpresent, the idea of high speed network processing has become important. Running this processing on a general purpose CPUs improves flexibility and reduces cost.
In this talk, we will discuss how the normal Linux network stack is limited in its' ability to process packets at a high throughput of 10Gb/s and above.
We will go over the inherent limitations of a CPU and then introduce DPDK; a framework for high performance networking on a CPU.
How does memory access, CPU scheduler, PCIe and other system components affect and determine network performance?
What does DPDK do inorder to overcome these bottlenecks ?
As an example, we will show a DPDK program running on the FABRIC research testbed, that does basic L2 forwarding, and we also will explain the main code sections.
Finally, we show a widely used packet generation tool written in DPDK, called DPDK-Pktgen and explain in brief how it can be used in experiments.

Bio:
Nishanth Shyamkumar is a Research Software Engineer at The Illinois Institute of Technology, Computer Science department. His current areas of research are in Distributed Systems, High performance networking and programmable networking.