Commit c6b94cf5 authored by Brewer, Wes's avatar Brewer, Wes
Browse files

Add in synthetic workload test for network -w network_test (see README.md for example)

parent 63fefdeb

README.md

+4 −0
Original line number Diff line number Diff line
@@ -89,6 +89,10 @@ get the datasets. To run a network simulation, use the following command: 


    raps run -f /opt/data/lassen/Lassen-Supercomputer-Job-Dataset --system lassen --policy fcfs --backfill firstfit --start '2019-08-22T00:00:00+00:00' -t 12h --arrival poisson --net



To simulate synthetic network tests:



    raps run --system lassen -w network_test --net -t 15m



## Snapshot of extracted workload data



To reduce the expense of extracting the needed data from the telemetry parquet files,

raps/sim_config.py

+2 −2
Original line number Diff line number Diff line
@@ -134,8 +134,8 @@ class SimConfig(RAPSBaseModel, abc.ABC): 
    """ Grab data from live system. """



    # Workload arguments (TODO split into separate model)

    workload: Literal['random', 'benchmark', 'peak', 'idle',

                      'synthetic', 'multitenant', 'replay', 'randomAI'] = "random"

    workload: Literal['random', 'benchmark', 'peak', 'idle', 'synthetic',

                      'multitenant', 'replay', 'randomAI', 'network_test'] = "random"



    """ Type of synthetic workload """

    multimodal: list[float] = [1.0]

raps/workloads/__init__.py

+3 −1
Original line number Diff line number Diff line
@@ -14,6 +14,7 @@ from .constants import JOB_NAMES, ACCT_NAMES, MAX_PRIORITY 
from .distribution import DistributionWorkload

from .live import continuous_job_generation

from .multitenant import MultitenantWorkload

from .network import NetworkTestWorkload

from .utils import plot_job_hist
@@ -51,7 +52,8 @@ class Workload( 
    BaseWorkload,

    DistributionWorkload,

    BasicWorkload,

    MultitenantWorkload

    MultitenantWorkload,

    NetworkTestWorkload

):

    """Final workload class with all workload types."""

    pass

raps/workloads/network.py

0 → 100644
+36 −0
Original line number Diff line number Diff line 


from raps.job import Job, job_dict



class NetworkTestWorkload:

    def network_test(self, **kwargs):

        """

        A synthetic workload to test network congestion.

        """

        config = kwargs.get('config', {})

        # High network traffic to trigger congestion

        # These values are per-node, and the network simulation sums them up

        # so we need to make them high enough to exceed the total network bandwidth

        net_tx = 1e12  # bytes

        net_rx = 1e12  # bytes



        job_info = job_dict(

            nodes_required=2,

            name="network-test-job",

            account="test",

            cpu_trace=[1],

            gpu_trace=[1],

            ntx_trace=[net_tx],

            nrx_trace=[net_rx],

            end_state='COMPLETED',

            id=1,

            priority=100,

            partition='partition',

            submit_time=0,

            time_limit=3600,

            start_time=0,

            end_time=3600,

            expected_run_time=3600,

            trace_quanta=20,

        )

        job = Job(job_info)

        return [job]