[dpdk-dev] [Bug 48] Unexpected performance regression since CVE-2018-1059 fix with vector path

[dpdk-dev] [Bug 48] Unexpected performance regression since CVE-2018-1059 fix with vector path

[bugzilla]

https://dpdk.org/tracker/show_bug.cgi?id=48

            Bug ID: 48
           Summary: Unexpected performance regression since CVE-2018-1059
                    fix with vector path
           Product: DPDK
           Version: 18.05
          Hardware: All
                OS: All
            Status: CONFIRMED
          Severity: normal
          Priority: Normal
         Component: vhost/virtio
          Assignee: dev@dpdk.org
          Reporter: maxime.coquelin@redhat.com
  Target Milestone: ---

A small performance drop was expected with CVE-2018-1059 fix integration.
While this expectation is verified for mast cases (less than 3% drop), a bigger
drop is seen when using vector path in guest with small packets.

The goal of this ticket is to see if this performance regression can be 
improved. 

For reference, below is a copy paste from Lei Yao mail reporting the issue:

During the 18.05-rc1 performance testing, I find this patch set will bring
slightly performance drop on mergeable and normal path, and big performance
drop on vector path. Could you have a check on this? I know this patch is 
important for security. Not sure if there is any way to improve the
performance.

Mergebale       
packet size     
64      0.80%
128     -2.75%
260     -2.93%
520     -2.72%
1024    -1.18%
1500    -0.65%

Normal  
packet size     
64      -1.47%
128     -7.43%
260     -3.66%
520     -2.52%
1024    -1.19%
1500    -0.78%

Vector  
packet size     
64      -8.60%
128     -3.54%
260     -2.63%
520     -6.12%
1024    -1.05%
1500    -1.20% 

CPU info: Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz
OS: Ubuntu 16.04

-- 
You are receiving this mail because:
You are the assignee for the bug.