You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Notice slower performance when using netmap on newer Linux kernels (e.g. 4.14, 4.18 on 4.19), when compared to much older kernels (e.g. 3.10.33). This seems especially true on slower CPUs.
For example, Intel i7-3770 CPU @ 3.40GHz gets 9.760Gbps when replaying bigFlows.pcap at full speed in 3.10.33 kernel. On a 4.14 kernel, that drops to about 6Gbps.
This is most likely due to Spectre and Meltdown security fixes. By turning off Spectre V2 (via spectre_v2=off kernel option) I was able to see improvement from 6 to 8Gbps. But still don't see all the performance I got with 3.10.33.
Most likely we have to live with this issue until Intel starts releasing hardware that addresses these security issues.
The text was updated successfully, but these errors were encountered:
This fix will yield the CPU while netmap is waiting
for buffers.
On my test system, this improved performance by only
60Mbps, not the 4Gbps performance loss identified
by this bug. Therefore, this is not the final fix
for this bug.
Notice slower performance when using netmap on newer Linux kernels (e.g. 4.14, 4.18 on 4.19), when compared to much older kernels (e.g. 3.10.33). This seems especially true on slower CPUs.
For example, Intel i7-3770 CPU @ 3.40GHz gets 9.760Gbps when replaying bigFlows.pcap at full speed in 3.10.33 kernel. On a 4.14 kernel, that drops to about 6Gbps.
This is most likely due to Spectre and Meltdown security fixes. By turning off Spectre V2 (via
spectre_v2=off
kernel option) I was able to see improvement from 6 to 8Gbps. But still don't see all the performance I got with 3.10.33.Most likely we have to live with this issue until Intel starts releasing hardware that addresses these security issues.
The text was updated successfully, but these errors were encountered: