stanford-futuredata gavel issues

stanford-futuredata / gavel

Code for "Heterogenity-Aware Cluster Scheduling Policies for Deep Learning Workloads", which appeared at OSDI 2020

MIT License

125 stars 31 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

Update sweep script to enable selecting between automatic and fixed-range sweeps

#47 santhnm2 closed 5 years ago
0
Set seeds according to user-provided input

#46 deepakn94 closed 5 years ago
0
Cleaned up notebook with latest packing results

#45 deepakn94 closed 5 years ago
0
Fix bug in isolated policy

#44 deepakn94 closed 5 years ago
0
Make FIFO deterministic and adhere to interval length

#43 santhnm2 closed 5 years ago
0
Fixed bug where jobs with 0 throughput received > 0 priority

#42 santhnm2 closed 5 years ago
0
Determine which workers should be active in a given round, taking into account both packed jobs and multi-GPU constraints

#41 deepakn94 closed 5 years ago
1
Clean up logging code

#40 santhnm2 closed 5 years ago
0
Code to fix computed utilization

#39 deepakn94 closed 5 years ago
0
Measure jobs in specified window

#38 santhnm2 closed 5 years ago
0
Update time resets to prioritize jobs with less service over time

#37 santhnm2 closed 5 years ago
0
Job completion times overestimated in round-based scheduling

#36 santhnm2 closed 5 years ago
1
De-duplicate functions for emulation

#35 santhnm2 closed 5 years ago
1
Implemented packing for rounds-based scheduling

#34 santhnm2 closed 5 years ago
0
Updated MSR traces and added traces with variable job arrival rate

#33 santhnm2 closed 5 years ago
0
Repository cleanup

#32 deepakn94 closed 5 years ago
1
Bugfixes to scheduling mechanism with rounds

#31 deepakn94 closed 5 years ago
0
New packed 'min total duration' policy

#30 deepakn94 closed 5 years ago
0
New 'minimize total duration' policy

#29 deepakn94 closed 5 years ago
1
Utilization measurements

#28 santhnm2 closed 5 years ago
1
Some fixes to get packing working

#27 deepakn94 closed 5 years ago
0
Reorganize traces/ directory to include tests

#26 deepakn94 closed 5 years ago
0
Start refactor of code

#25 deepakn94 closed 5 years ago
0
Incorporate packing in ideal

#24 deepakn94 closed 5 years ago
0
Bug fixes and performance optimizations for packing

#23 santhnm2 closed 5 years ago
1
Determine the ideal trace duration / average JCT for a trace

#22 deepakn94 closed 5 years ago
0
Refactor scheduler_with_rounds into scheduler.py

#21 deepakn94 closed 5 years ago
0
Add scale_factor in trace

#20 deepakn94 closed 5 years ago
0
Schedule in rounds (instead of one-at-a-time)

#19 deepakn94 closed 5 years ago
1
Added new workloads, online throughput profiling, and accurate emulator

#18 santhnm2 closed 5 years ago
1
Refactor messy code in scheduler

#17 deepakn94 closed 5 years ago
0
Compute KL divergence between true distribution and allocated distribution

#16 deepakn94 closed 4 years ago
0
Various scheduler fixes

#15 deepakn94 closed 5 years ago
1
Use pre-measured throughputs when command line arg available

#14 deepakn94 closed 5 years ago
0
Support emulation of a trace

#13 deepakn94 closed 5 years ago
1
Allow scheduler to terminate workers

#12 santhnm2 closed 5 years ago
0
Allow commands in trace to be run at specified points in time

#11 deepakn94 closed 5 years ago
0
Refactor code in scheduler.py

#10 deepakn94 closed 5 years ago
0
Reset time_run_so_far whenever an event (add_job, remove_job, add_worker) occurs

#9 deepakn94 closed 5 years ago
0
Use worker type instead of worker ID for scheduling

#8 deepakn94 closed 5 years ago
0
Changes to support running multiple applications on a single worker c…

#7 deepakn94 closed 5 years ago
0
Improve error handling

#6 deepakn94 closed 4 years ago
0
Shut down workers when `run_scheduler_with_trace.py` is done

#5 deepakn94 closed 5 years ago
2
Time multiplex instead of step multiplex

#4 deepakn94 closed 5 years ago
0
Changes to incorporate multiple workers into scheduler

#3 deepakn94 closed 5 years ago
0
[WIP] Changes to support multiple workers

#2 deepakn94 closed 5 years ago
0
Scheduler implementation that can multiplex across applications according to computed allocation

#1 deepakn94 closed 5 years ago
0