Hi Hongjun, Yes, sure. I will try the patch and share my results as soon as possible.
Regards, Abeeha Aqeel Network Design Engineer Xflow Research Inc. +923245062309 (GMT+5) abeeha.aq...@xflowresearch.com www.xflowresearch.com From: Ni, Hongjun Sent: Tuesday, May 7, 2019 6:40 PM To: Abeeha Aqeel; Singh, Jasvinder Cc: b...@xflowresearch.com; vpp-dev@lists.fd.io Subject: Re: [vpp-dev] HQoS Hi Abeeha, There is a patch submitted to fix the API change. https://gerrit.fd.io/r/#/c/16839/ Could you help to give it a try? Thanks, Hongjun From: vpp-dev@lists.fd.io [mailto:vpp-dev@lists.fd.io] On Behalf Of Abeeha Aqeel Sent: Monday, May 6, 2019 6:32 PM To: Ni, Hongjun <hongjun...@intel.com> Cc: b...@xflowresearch.com; vpp-dev@lists.fd.io Subject: Re: [vpp-dev] HQoS Hi Hongjun, I have tested the HQoS Plugin with iperf3. Below is a simple topology I have implemented: First, I tested the default profile (profile 0) provided by VPP and assigned it to all 4096 pipes (users) which should give approximately 2MB (10G/4096) to each user. It worked fine for 5 users i.e. each user was assigned 2MB but for an increased number of users it didn’t work the same way. I changed the token bucket rate, token bucket size and tc_period parameters in the hqos.c code which didn’t give me any concrete results. I have attached results for 10 clients with different combinations of token bucket size, token bucket rate and tc_period. Any ideas what the problem might be ? Also, in what proportion should the tb_rate, tb_size , tc_rate and tc_period be assigned? S# Rate (Bytes per second) Token Bucket Size (Bytes) TC Period (ms) Total no of PPPoE sessions Expected Result (Cumulative average bandwidth in Mbps) Actual Result (Cumulative average bandwidth in Mbps) Comments Subport Pipe Subport Pipe Subport Pipe 1 1250000000 305175 1000000 1000000 10 40 10 24.4 8.1 vpp remains stable 2 1250000000 305175 1000000 1000000 10 60 10 24.4 12.72 vpp remains stable 3 1250000000 305175 1000000 1000000 10 100 10 24.4 11.4 vpp remains stable 4 1250000000 305175 1000000 10000000 10 40 10 24.4 12.1 vpp remains stable 5 1250000000 305175 10000000 10000000 10 40 10 24.4 12.03 vpp remains stable 6 1250000000 305175 10000000 10000000 10 100 10 24.4 2.21 vpp crashes after a few seconds 7 1250000000 305175 1000000 1000000 10 60 10 24.4 12.1 vpp remains stable Regards, Abeeha Aqeel Network Design Engineer Xflow Research Inc. +923245062309 (GMT+5) abeeha.aq...@xflowresearch.com www.xflowresearch.com From: Ni, Hongjun Sent: Monday, May 6, 2019 12:52 PM Subject: RE: [vpp-dev] HQoS Hi Abeeha, For downstream bandwidth limiting, we leveraged HQos plugin in OpenBRAS solution. In our previous integration test, it could support 64K subscribers with HQos. Thanks, Hongjun From: vpp-dev@lists.fd.io [mailto:vpp-dev@lists.fd.io] On Behalf Of Abeeha Aqeel Sent: Monday, May 6, 2019 3:28 PM To: Ni, Hongjun <hongjun...@intel.com> Cc: b...@xflowresearch.com; vpp-dev@lists.fd.io Subject: [vpp-dev] HQoS Hi Hongjun, I have been trying to implement downstream bandwidth limiting using HQoS plugin in vpp. It works fine for a certain number of clients (less than 5) but doesn’t assign proper bandwidth for larger number of clients. Can you please elaborate which method is being used in the OpenBRAS solution for band limiting traffic? Are you using the token-bucket algorithm as well? Regards, Abeeha Aqeel Network Design Engineer Xflow Research Inc. +923245062309 (GMT+5) abeeha.aq...@xflowresearch.com www.xflowresearch.com
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#12961): https://lists.fd.io/g/vpp-dev/message/12961 Mute This Topic: https://lists.fd.io/mt/31517544/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-
