High speed tuning for SimpleNetworkRecipe

[jtluka]

Description

This PR contains various changes that are required to get the best performance from high speed networks such as 40gbit and 100gbit NICs/networks.

The changes include:

• new mixins or mixins that allows per-host/per-device configuration of settings available in existing mixins (device IRQ pinning, adaptive coalescing, etc.).
• support for multi cpu pinning of IperfClient/Server
• reusability of test wide configuration of SimpleNetworkRecipe
• FlowMeasurementMultiCpupinGenerator that provides separate parameters for generator/receiver

All these changes should not break existing recipes and should be transparent - I plan to do a test in internal labs to be sure.

Tests

Test a tuned UDP 100gb scenario: J:8013224 (works fine, tx: 100gbit/s, rx: 51 gbit/s) Test original SimpleNetworkRecipe: J:8019807

Reviews

@olichtne

[jtluka]

An issue was seen in internal run of SimpleNetworkRecipe with baseline evaluation, job 8019807. This needs to be resolved first.

Update:

The issue is due to downstream code not adapted to the changes in this MR, in particular split of cpupin to generator_cpupin and receiver_cpupin.

I've adapted downstream code and ran another test job 8019994.

Update: the job with baseline evaluation passed after fix.

[olichtne]

i think this looks good now, so @jtluka if you're ready, feel free to rebase, squash and merge this

[jtluka]

Rebased/squashed and moved the commits a bit to have it in chronological sense.

New test jobs: J:8020936 J:8020935