Hi Chungan, Replies in-line below.
Thanks, Billy From: Lee, Chunghan [mailto:[email protected]] Sent: Monday, December 12, 2016 5:42 AM To: O Mahony, Billy <[email protected]>; [email protected] Cc: Gray, Mark D <[email protected]>; [email protected] Subject: RE: [OVSNFV] Sharing the priority path result Hi Billy, Thank you for sharing the results. I can understand the results. Particularly, the result (10to1Pri at 120%Max) is very interested. In my understanding, the latency at the overloaded condition (120%Max) is largely increased regardless of Priority Queue configuration. [[BO'M]] There is no priority queue in this test. This is simply a baseline to show the current latency performance of the system. When I have completed the priority path code I will redo the tests to show latency performance when a priority does exist for high priority traffic. But, why the range of latency (20-400) is exist with 40%Max or 80%Max? What is a major cause of the range? # PMD cannot forward packets correctly and increase the latency? [[BO'M]] That is a very good question! Interesting further work could be to discover where and why latency is introduced. For instance do high latency packets come in groups/clumps or are they randomly distributed. One thing to remember is that the y-axis is a log scale so it is just a few packets in every 100,000 that suffer latencies in the 20-400 us range. What is 800-INF? I’m so sorry that I cannot understand this caption at the X-axis. [[BO'M]] This ‘latency bucket’ counts packets whose latency is over 800 micro seconds. The value is expressed as a ratio of the number of packets in the latency bucket to the total number of packets for the traffic type. So a figure of .000001 would mean that 1 packet in every million had a latency of over 800us. In 120%Max, the ratio of packet loss is larger than 0.002 ? [[BO'M]] Yes. Part of was to find the maximum throughput of the DUT with the particular packet ratio’s required. So by definition at 100% throughput the DUT will already be losing around .002% of packets. So adding an extra 20% to the offered load will result in much more packet loss. Let me reconfirm the entire experiment configuration. The traffic latency is same to the packet latency? [[BO'M]] We are using a h/w traffic generator (IXIA) to measure the latencies so it is full end to end latency. The latency is one way and it measured ptkgen. # I found the term (Unidirectional Traffic…). But, I didn’t understand the entire topology and other configurations. Is there more detail information to understand the configuration file, such as yardstick (YAML file)? If YES, could you please share the detail information? [[BO'M]] If you look at the slides here you will see the topology of “PVP” test. http://prezi.com/tlriipoafn5_/?utm_campaign=share&utm_medium=copy&rc=ex0share Regards, Chunghan Lee From: O Mahony, Billy [mailto:[email protected]] Sent: Monday, December 05, 2016 11:08 PM To: Lee, Chunghan/李 忠翰; [email protected]<mailto:[email protected]> Cc: Gray, Mark D; [email protected]<mailto:[email protected]> Subject: RE: [OVSNFV] Sharing the priority path result Hi Chunghan, Pls find attached. From: Lee, Chunghan [mailto:[email protected]] Sent: Monday, December 5, 2016 1:11 AM To: [email protected]<mailto:[email protected]> Cc: O Mahony, Billy <[email protected]<mailto:[email protected]>>; Gray, Mark D <[email protected]<mailto:[email protected]>>; [email protected]<mailto:[email protected]>; Lee, Chunghan <[email protected]<mailto:[email protected]>> Subject: [OVSNFV] Sharing the priority path result Hi Billy, At the previous meeting, you, Billy, introduced the priority path result. If no problem, could you share the result ? I did not understand the result fully. In my memory, Tom also required the sharing. Regards, Chunghan Lee
_______________________________________________ opnfv-tech-discuss mailing list [email protected] https://lists.opnfv.org/mailman/listinfo/opnfv-tech-discuss
