dpdk17

Format:
This session will have both presentation and demo.
(In the abstract, there are figures. So I e-mailed you the abstract with figures)
Presentation Abstract:

DPDK – Multiple Architectures:
While all architectures, ARM, POWER, X86, run DPDK, not only have architectural differences, they also have variance in platform settings (like NUMA, Core affinity and memory channels).

To have apple to apple comparison, developers need a common ground of base level metrics.
That common ground is ability to identify the basic DPDK building block of importance (as well as relevance to the work load) e.g., producer/consumer rings and measure the cycle cost associated with basic operation like enque/dequeing – bulk versus single .

The best kept secret here is - DPDK has Micro benchmark performance tests for various key elements (eg., rings, mbufs, mpools, ports, threads) doing exactly that. It is best kept secret because one will be surprised to know how many developers are looking for such tool and very few know it exists in DPDK.

I will point the location in DPDK and show how these test groups can be run. Pointing these test groups to developers at DPDK Summit would ensure developers speaking common metrics across architectures.


DPDK– Multiple MicroArchitectures:
Not only across architectures, even within same architecture, microarchitecture changes effect these numbers. To illustrate the point, I will present non-inclusive cache – first time introduced in latest Scalable processor architecture and its ramifications in Inter processor communication.


Demo with latest processor & platform

To drive home the point on the benefit of DPDK micro benchmark performance tests, I will bring platform to the summit and run DPDK 17.08 tests on latest Scalable processor SoC platform and show the cycle cost of IPC between siblings and non-siblings and chart out a graph – as shown below. This will benefit the developers by profiling their platform at gross root level cycle cost of most important element and basic operation.