Commit 00382592 authored by Brewer, Wes's avatar Brewer, Wes
Browse files

Add Frontier aging policy

parent 7e4a960f

main.py

+5 −4
Original line number Diff line number Diff line
@@ -82,8 +82,8 @@ if args.replay: 
        extracted_date = "Date not found"

        DIR_NAME = create_casename()



    # Read either npz file or telemetry parquet files

    if args.replay[0].endswith(".npz"):

    # Read telemetry data

    if args.replay[0].endswith(".npz"): # read .npz file

        print(f"Loading {args.replay[0]}...")

        jobs = td.load_snapshot(args.replay[0])
@@ -98,9 +98,10 @@ if args.replay: 
                job['requested_nodes'] = None

                job['submit_time'] = next_arrival(1 / config['JOB_ARRIVAL_TIME'])



    else:

    else: # custom data loader 

        print(*args.replay)

        jobs = td.load_data(args.replay)

        for job in jobs: job['priority'] = sc.policy.aging_boost(job['nodes_required'])

        td.save_snapshot(jobs, filename=DIR_NAME)



    # Set number of timesteps based on the last job running which we assume
@@ -113,7 +114,7 @@ if args.replay: 
    print(f'Simulating {len(jobs)} jobs for {timesteps} seconds')

    time.sleep(1)



else:

else: # synthetic jobs

    wl = Workload(config)

    jobs = getattr(wl, args.workload)(num_jobs=args.numjobs)

raps/policy.py

+9 −0
Original line number Diff line number Diff line
@@ -23,6 +23,15 @@ class Policy: 
        else:

            raise ValueError(f"Unknown policy type: {self.policy_type}")



    def aging_boost(self, nnodes):

        """Frontier aging policy"""

        if nnodes > 5645:

            return 8

        elif nnodes > 1882:

            return 4

        else:

            return 0 



    def find_backfill_job(self, queue, num_free_nodes, current_time):

        """ This implementation is based on pseudocode from Leonenkov and Zhumatiy.

            "Introducing new backfill-based scheduler for slurm resource manager."