IX Open-source version 1.0 { Deployment and Evaluation Guide
George Prekas, Adam Belay, Mia Primorac, Ana Klimovic, Samuel Grossman, Marios Kogias, Christos Kozyrakis, Edouard Bugnion · 2016
AbstractThis Technical Report provides the deployment and evaluation guide of the ix dataplane operatingsystem, as of its rst open-source release on May 27, 2016. To facilitate the reproduction of ourresults, we include in this report the precise steps needed to install, deploy and con gure ix and itsworkloads. We reproduce all benchmarks previously published in two peer-reviewed publications atOSDI ’14 [3] and SoCC ’15 [6] using this up-to-date, open-source code base. Revision history Version Date Summaryv1.1 April 28, 2017 Update Fig. 5, Fig 6, and Table 1. In v1.0 the cor-responding benchmark was executed on an emptymemcached database by mistake.v1.0 May 27, 2016 Initial release. 1 Introduction The IX dataplane operating system project started in 2013 as a collaboration between researchersat Stanford and EPFL. The design and implementation of the dataplane, which combines low latencyand high throughput, was published at OSDI in 2014 [3]. The design and implementation of the controlplane, which additionally ensures high system eciency, was published at SoCC in 2015 [6].Like all software projects, IX is in constant evolution. In recent months, the focus was to increasethe robustness, debuggability and extensibility of the system.As we release the code base in open-source for the rst time, this report documents the four necessarycomponents required to reproduce our research results: the IX software: this includes instructions to build and deploy IX, including its dependencies:Dune [2] and DPDK [4]; the benchmarks: the full set of benchmarks that were used in prior publications. the experimental setup: this includes a description of the hardware used for those experimentsas well as the complete description of the host Linux con guration. the experimental results: a complete re-evaluation of IX on all benchmarks, performed withthe open-source code base. The results are generally consistent with the published numbers.This work was funded by DARPA CRASH (under contract #N66001-10-2-4088), a Google researchgrant, the Stanford Experimental Datacenter Lab, the Microsoft-EPFL Joint Research Center, NSFgrant CNS-1422088, and a VMware grant. George Prekas was supported by a Google Graduate ResearchFellowship and Adam Belay by a VMware Graduate Fellowship.1