[ovs-discuss] Re： [HELP] Question about Performance issues in high concurrency scenarios

[txfh2007]

Hi Yanqin && all:
    I have studied this ppt && vedio, and  tried in my own environment. The result is in 100K mac-ip PVP test scenario, when EM_FLOW_HASH_SHIFT == 17(emc entry num is 130K+), the performance can still maintain(a little decline than 10K mac-ip scenario). My formal conclusion is wrong as I didn't notice that if I generate 100K mac-ip flows , then the mac-table-entry num in ovs bridge should be enlarge simultaneously(I use the auto learn flow entries)

   I also tested scenario that smc is enabled and emc disabled. My result is pure smc performance is slightly not good as emc.(100K mac-ip flows, pkt size is 64 bytes, 4pmd lcores , 10G nic , PVP test, smc can reach 3.7G throughput, latency is 56us. For emc throughput is 4G , latency 51us )

   My question is: should emc entry num get enlarged ? Are there any drawbacks if emc entry is larger than 100K ?







------------------------------------------------------------------
发件人：Yanqin Wei (Arm Technology China) <Yanqin.Wei at arm.com>
发送时间：2019年3月14日(星期四) 10:05
收件人：txfh2007 <txfh2007 at aliyun.com>; Flavio Leitner <fbl at sysclose.org>
抄　送：ovs-discuss <ovs-discuss at openvswitch.org>
主　题：RE: [ovs-discuss] Re： [HELP] Question about Performance issues in high concurrency scenarios


!--  @font-face{font-family:Cambria Math;panose-1:2 4 5 3 5 4 6 3 2 4;}{font-family:DengXian;panose-1:2 1 6 0 3 1 1 1 1 1;}{font-family:Calibri;panose-1:2 15 5 2 2 2 4 3 2 4;}{font-family:\@DengXian;panose-1:2 1 6 0 3 1 1 1 1 1;}{font-family:Tahoma;panose-1:2 11 6 4 3 5 4 4 2 4;}{font-family:Microsoft YaHei;panose-1:2 11 5 3 2 2 4 2 2 4;}{font-family:\@Microsoft YaHei;}p.MsoNormal, li.MsoNormal, div.MsoNormal{margin:.0in;margin-bottom:.0pt;font-size:11.0pt;font-family:Calibri,sans-serif;}a:link, span.MsoHyperlink{mso-style-priority:99;color:#0563c1;text-decoration:underline;}a:visited, span.MsoHyperlinkFollowed{mso-style-priority:99;color:#954f72;text-decoration:underline;}p.msonormal0, li.msonormal0, div.msonormal0{mso-style-name:msonormal;mso-margin-top-alt:auto;margin-right:.0in;mso-margin-bottom-alt:auto;margin-left:.0in;font-size:11.0pt;font-family:Calibri,sans-serif;}span.EmailStyle19{mso-style-type:personal-reply;font-family:Calibri,sans-serif;color:windowtext;}.MsoChpDefault{mso-style-type:export-only;font-family:Calibri,sans-serif;}@page WordSection1{size:8.5in 11.0in;margin:1.0in 1.0in 1.0in 1.0in;}div.WordSection1{page:WordSection1;}-->Hi,
 
There is a interesting topic in Open vSwitch 2018 Fall Conference maybe helpful for you. It shows EMC/SMC behave different performance in high concurrency case.
For your case, SMC may be more helpful.
 
http://www.openvswitch.org/support/ovscon2018/5/1330-theurer.pdf
 
 
 
From: ovs-discuss-bounces at openvswitch.org <ovs-discuss-bounces at openvswitch.org>On Behalf Of txfh2007 via discuss
Sent: Thursday, March 14, 2019 9:00 AM
To: Flavio Leitner <fbl at sysclose.org>
Cc: ovs-discuss <ovs-discuss at openvswitch.org>
Subject: [ovs-discuss] Re： [HELP] Question about Performance issues in high concurrency scenarios
 
Thanks，Flavio:

    So there are two questions: 

    1. Is dpcls lookup More efficiency than EMC?

    2. If I just want to enlarge emc entry num by change EM_FLOW_HASH_SHIFT value, are there any drawbacks ?

 

 

------------------------------------------------------------------

发件人：Flavio Leitner <fbl at sysclose.org>

发送时间：2019年3月14日(星期四) 04:05

收件人：txfh2007 <txfh2007 at aliyun.com>

抄　送：ovs-discuss <ovs-discuss at openvswitch.org>

主　题：Re: [ovs-discuss] [HELP] Question about Performance issues in high concurrency scenarios

 

On Tue, Mar 05, 2019 at 03:38:43PM +0800, txfh2007 via discuss wrote:
> Hi everyone:    
>      I have test ovs-dpdk PVP performance for several days . And I
>      have found in high concurrency scenarios(eg. 100K mac-ip
>      streams), the ovs-dpdk performance gets significant decline.
>      And one cause of performance decline is the emc entry num
>      limitation, is that right? If I just increase emc entry num,
>      will other problems occur ? Are there any suggession helping to
>      get performance improvement under high concurrency scenarios?
 

Perhaps you could consider disabling EMC if its size is not enough.
fbl
 



IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.