Internode optimizer rework and integration with dashboard #33
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Task-migration still fails but much less frequently
- Dashboard displays power as a fraction of the current power consumption over the maximum power consumption. - It uses a model to generate these values. That can be inaccurate. - This commit, reads the real power consumption but does not supply a maximum power consumption.
easier integration with dashboard
Do a final run once NMD converges to make sure that the scores of the simplices are not stale
Removed freq_idx domain reduction
- If we re-enable that feature we probably have to add a map that
converts a "fake" freq_idx to the actual one
…ency" - This is only the case when CPUFreq is used - I've also modified the dashboard
Added a virtual to physical node dictionary. INO_NMD assumes that nodes are in sequential order while it is making decision on which nodes to use but once it has made its choices it makes sure that it uses only nodes which are working
- Though it still uses "threads" and "freq_idx" as knob names the only
restriction is that these values must be dicrete (i.e integers)
- Expects that step() contains the actual measured values of "the knob
set".
- For example, the scheduler may not always manage to resume/suspend
the number of threads that the local-optimizer suggests
- Added a maximum region that can be searched for new knob_set
alternatives so that the spirit of the optimization process is kept
sort of intact
- We can set CPU governor, retrieve and set CPU frequency
… measurement information that's cached
…ons for the NelderMead algorithm
- NmdGeneric accepts a score-function that's a parameter - Reverted default score function to speed/efficiency/power (dashboard) - Fixed a couple of small bugs in nmsimplex_bbincr (for use with legacy global and local NMD optimizers which rely on time/threads/power)
(re-enable #defines to get the functionality back)
previous commit was incomplete ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Optimized legacy NMD implementation
Implemented Generic NMD algorithm for non-continuous optimization search spaces (e.g. number of threads, CPU frequency, number of nodes, etc)
Added a direct replacement to tuner (which can be toggled via ALLSCALE_TUNER env param). It uses the speed/efficiency/power utility function