Dear All, We have continued our investigation, but now we have considered OVS 2.5.1 with *DPDK driver* 16.04 and conducted some measurements with as the same use cases and traces as before (bridge, l2-switch, l3-router, load_balancer_1,10,100).
We also uploaded it to NFPA's site. Now, if you go to http://nfpa.tmit.bme.hu/index.php?page=browse and select *ovs* as VNF name and *2.5.1* as version, you will also have *dpdk* as vnf driver. Those are the fresh results. For some quick snippets, check the links below: https://goo.gl/fmZpx9 - OVS bridge performance in terms of used cores (1, 3, 6) https://goo.gl/mT5qOV - OVS l2-switch performance as a function of different number of flow rules (100,1000,10.000,100.000) using 1 core https://goo.gl/ICTTIY - the same but with 4 cores https://goo.gl/QZP1vt - OVS l3-router performance as a function of different number of flow rules (100,1000,10.000,100.000) using 1 core https://goo.gl/3ObzPU - the same but with 4 cores https://goo.gl/IdMoHd - Load_balancer_100 use-case with 1000, 10.000, and 100.000 incoming flows using 1 core. Feel free to browse and compare. -- Cheers, Barry On Sun, Apr 9, 2017 at 5:08 PM, L C <[email protected]> wrote: > Dear All, > > I try to keep up with the things going on on this mailing list, and I hope > I don't ask/share any outdated information. This is also my first time of > posting something to here. > > So, we (as might many others) were curious about the data plane throughput > of the Open vSwitch. > We found a website offering an open-source benchmarking tool with a lot of > useful other stuffs (e.g., traces, pipelines), called NFPA ( > http://nfpa.tmit.bme.hu). > > We have downloaded the tool and evaluated OVS (v2.5.1) over our testbed > (Intel Xeon, XL710...), but currently only the kernel data path is done. > As the website offers sharing mechanism, we put our results there. > > we evaluated the predefined use cases: bridge, l2-switch, l3-router, > load_balancer_1,10,100 (http://nfpa.tmit.bme.hu/index.php?page=usecases) > > > here is a snippet if anyone interested (shortened url to NFPA): http:// > goo.gl/p7S0Ga > > Because of 40G interface, the plots are not really talky, even if you zoom > into them, so use the table views instead, it might provide you the exact > numbers. The shared link is only for traffic trace where 100 different > flows were in the traces, but you can simply browse other cases on the > website. > > It seems that if you choose version 2.5.1 on the browsing subpage, you > will only find our freshly uploaded measurements :) > > > Best regards, > Barrred Kola > >
_______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
