jobs.yml

# Copyright (c) Meta Platforms, Inc. and affiliates.
#
# This source code is licensed under the MIT license found in the
# LICENSE file in the root directory of this source tree.
- benchmark: oss_performance_mediawiki
  name: oss_performance_mediawiki
  description: Default run for oss_performance_mediawiki
  args:
    - '-r/usr/local/hphpi/legacy/bin/hhvm'
    - '-nnginx'
    - '-L {load_generator}'
    - '-s {lg_path}'
    - '-U {auto_fix_ulimit}'
    - '--'
    - '--mediawiki'
    - '--client-duration={duration}'
    - '--client-timeout={timeout}'
    - '--run-as-root'
    - '--i-am-not-benchmarking'
    - '{extra_args}'
  vars:
    - 'load_generator=wrk'
    - 'lg_path=benchmarks/oss_performance_mediawiki/wrk/wrk'
    - 'duration=10m'
    - 'timeout=11m'
    - 'auto_fix_ulimit=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'oss-performance/perf.data'

- benchmark: oss_performance_mediawiki
  name: oss_performance_mediawiki_mini
  description: Default run for oss_performance_mediawiki
  execute_cmd_from_file: true
  args:
    - '-r/usr/local/hphpi/legacy/bin/hhvm'
    - '-nnginx'
    - '-L {load_generator}'
    - '-s {lg_path}'
    - '-T {temp_dir}'
    - '-U {auto_fix_ulimit}'
    - '-p'
    - '--'
    - '--mediawiki'
    - '--client-duration={duration}'
    - '--client-timeout={timeout}'
    - '--run-as-root'
    - '--i-am-not-benchmarking'
    - '--shorten-health-check'
    - '--skip-single-request-warmup'
    - '--skip-sleep-between-warmups'
    - '--hhvm-extra-arguments="-vEval.JitRetranslateAllSeconds={translation_interval_seconds}"'
    - '--num-multi-req-warmups={num_multi_req_warmups}'
    - '--no-load-if-pending-translate'
    - '--first-multi-warmup-duration={first_multi_warmup_duration}'
    - '--subseq-multi-warmup-duration={subseq_multi_warmup_duration}'
    - '--load-gen-seed={load_generator_seed}'
    - '{extra_args}'
  vars:
    - 'load_generator=wrk'
    - 'lg_path=benchmarks/oss_performance_mediawiki/wrk/wrk'
    - 'duration=4s'
    - 'timeout=60s'
    - 'translation_interval_seconds=5'
    - 'first_multi_warmup_duration=25' # long enough to get a high score
    - 'subseq_multi_warmup_duration=5'
    - 'temp_dir=default_no_temp_dir'
    - 'num_multi_req_warmups=-1' # -1 means the warmup status will define the number of iterations
    - 'load_generator_seed=1000'
    - 'auto_fix_ulimit=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'oss-performance/perf.data'

- benchmark: oss_performance_mediawiki
  name: oss_performance_mediawiki_mlp
  description: Tuned +MLP run for oss_performance_mediawiki
  args:
    - '-r/usr/local/hphpi/legacy/bin/hhvm'
    - '-nnginx'
    - '-L {load_generator}'
    - '-s {lg_path}'
    - '-R{scale_out}'
    - '-c{client_threads}'
    - '-t{server_threads}'
    - '-m{memcache_threads}'
    - '-U {auto_fix_ulimit}'
    - '--'
    - '--mediawiki-mlp'
    - '--client-duration={duration}'
    - '--client-timeout={timeout}'
    - '--run-as-root'
    - '--i-am-not-benchmarking'
    - '{extra_args}'
  vars:
    - 'load_generator=wrk'
    - 'lg_path=benchmarks/oss_performance_mediawiki/wrk/wrk'
    - 'scale_out=0'
    - 'client_threads=0'
    - 'server_threads=0'
    - 'memcache_threads=0'
    - 'duration=10m'
    - 'timeout=11m'
    - 'auto_fix_ulimit=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'oss-performance/perf.data'

- benchmark: oss_performance_mediawiki
  name: oss_performance_mediawiki_mlp_no_jit
  description: Tuned +MLP run for oss_performance_mediawiki
  args:
    - '-r/usr/local/hphpi/legacy/bin/hhvm'
    - '-nnginx'
    - '-L {load_generator}'
    - '-s {lg_path}'
    - '-R{scale_out}'
    - '-c{client_threads}'
    - '-t{server_threads}'
    - '-m{memcache_threads}'
    - '-U {auto_fix_ulimit}'
    - '--'
    - '--mediawiki-mlp'
    - '--client-duration={duration}'
    - '--client-timeout={timeout}'
    - '--run-as-root'
    - '--no-jit'
    - '--i-am-not-benchmarking'
    - '{extra_args}'
  vars:
    - 'load_generator=wrk'
    - 'lg_path=benchmarks/oss_performance_mediawiki/wrk/wrk'
    - 'scale_out=0'
    - 'client_threads=0'
    - 'server_threads=0'
    - 'memcache_threads=0'
    - 'duration=10m'
    - 'timeout=11m'
    - 'auto_fix_ulimit=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'oss-performance/perf.data'

- benchmark: oss_performance_mediawiki
  name: oss_performance_mediawiki_mem
  description: Tuned +(MLP+LambdaChase) run for oss_performance_mediawiki
  args:
    - '-r/usr/local/hphpi/legacy/bin/hhvm'
    - '-nnginx'
    - '-L {load_generator}'
    - '-s {lg_path}'
    - '-R{scale_out}'
    - '-c{client_threads}'
    - '-t{server_threads}'
    - '-m{memcache_threads}'
    - '-U {auto_fix_ulimit}'
    - '--'
    - '--mediawiki-mem'
    - '--client-duration={duration}'
    - '--client-timeout={timeout}'
    - '--run-as-root'
    - '--i-am-not-benchmarking'
    - '{extra_args}'
  vars:
    - 'load_generator=wrk'
    - 'lg_path=benchmarks/oss_performance_mediawiki/wrk/wrk'
    - 'scale_out=0'
    - 'client_threads=0'
    - 'server_threads=0'
    - 'memcache_threads=0'
    - 'duration=10m'
    - 'timeout=11m'
    - 'auto_fix_ulimit=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'oss-performance/perf.data'

- benchmark: django_workload
  name: django_workload_default
  description: Default run for django-workload
  tee_output: true
  roles:
    clientserver:
      args:
        - '-r clientserver'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-c {db_addr}'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'db_addr'
        - 'duration=5M'
        - 'iterations=7'
        - 'reps=0'
        - 'interpreter=cinder'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
        - 'use_jit=1'
        - 'skip_datagen=0'
    db:
      args:
        - '-r db'
        - '-b {bind_ip}'
        - '--thrift-server-workers {thrift_server_workers}'
      vars:
        - 'bind_ip=default'
        - 'thrift_server_workers=0'
    server:
      args:
        - '-r server'
        - '-c {db_addr}'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'db_addr'
        - 'interpreter=cinder'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
        - 'use_jit=1'
        - 'skip_datagen=0'
    standalone:
      args:
        - '-r standalone'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-c 127.0.0.1'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--thrift-server-workers {thrift_server_workers}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'duration=5M'
        - 'iterations=7'
        - 'reps=0'
        - 'interpreter=cinder'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
        - 'use_jit=1'
        - 'thrift_server_workers=0'
        - 'skip_datagen=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/django_workload/django-workload/client/perf.data'
          - 'benchmarks/django_workload/django-workload/django-workload/*.log'
          - 'benchmarks/django_workload/django-workload/django-workload/log_load_balancer'
          - 'benchmarks/django_workload/django-workload/client/*.log'
          - 'benchmarks/django_workload/*.log'
          - '/tmp/wrk_out_*'
          - 'benchmarks/django_workload/breakdown.csv'

- benchmark: django_workload
  name: django_workload_mini
  description: A short version of django-workload, where we can load the database from a snapshot
  tee_output: true
  roles:
    standalone:
      args:
        - '-r standalone'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-c 127.0.0.1'
        - '-S'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
      vars:
        - 'duration=1M'
        - 'iterations=1'
        - 'reps=1000'
        - 'interpreter=cpython'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/django_workload/django-workload/client/perf.data'
          - 'benchmarks/django_workload/django-workload/django-workload/*.log'
          - 'benchmarks/django_workload/django-workload/django-workload/log_load_balancer'
          - 'benchmarks/django_workload/django-workload/client/*.log'
          - 'benchmarks/django_workload/*.log'
          - '/tmp/wrk_out_*'
          - 'benchmarks/django_workload/breakdown.csv'


- benchmark: django_workload
  name: django_workload_custom
  description: Django-workload benchmark with custom parameters
  tee_output: true
  roles:
    clientserver:
      args:
        - '-r clientserver'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-c {db_addr}'
        - '-w {server_workers}'
        - '-x {client_workers}'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'duration=5M'
        - 'iterations=7'
        - 'reps=0'
        - 'db_addr'
        - 'server_workers'
        - 'client_workers'
        - 'interpreter=cpython'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
        - 'use_jit=0'
        - 'skip_datagen=0'
    client:
      args:
        - '-r client'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-x {client_workers}'
        - '-z {server_addr}'
      vars:
        - 'duration=5M'
        - 'iterations=7'
        - 'reps=0'
        - 'client_workers'
        - 'server_addr'
    server:
      args:
        - '-r server'
        - '-c {db_addr}'
        - '-w {server_workers}'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'db_addr'
        - 'server_workers'
        - 'interpreter=cpython'
        - 'use_async=1'
        - 'base_port=16668'
        - 'stats_port=8001'
        - 'use_jit=0'
        - 'skip_datagen=0'
    db:
      args:
        - '-r db'
        - '-y {concur_writes}'
        - '-b {bind_ip}'
        - '--thrift-server-workers {thrift_server_workers}'
      vars:
        - 'concur_writes'
        - 'bind_ip=default'
        - 'thrift_server_workers=0'
    standalone:
      args:
        - '-r standalone'
        - '-d {duration}'
        - '-i {iterations}'
        - '-p {reps}'
        - '-l ./siege.log'
        - '-s urls.txt'
        - '-c 127.0.0.1'
        - '-w {server_workers}'
        - '-x {client_workers}'
        - '--interpreter {interpreter}'
        - '--use-async {use_async}'
        - '--base-port {base_port}'
        - '-T {stats_port}'
        - '--use-jit {use_jit}'
        - '--thrift-server-workers {thrift_server_workers}'
        - '--skip-datagen {skip_datagen}'
      vars:
        - 'server_workers'
        - 'client_workers'
        - 'duration=5M'
        - 'iterations=7'
        - 'reps=0'
        - 'interpreter=cinder'
        - 'use_async=1'
        - 'base_port=16667'
        - 'stats_port=8001'
        - 'use_jit=1'
        - 'thrift_server_workers=0'
        - 'skip_datagen=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/django_workload/django-workload/client/perf.data'
          - 'benchmarks/django_workload/django-workload/django-workload/*.log'
          - 'benchmarks/django_workload/django-workload/django-workload/log_load_balancer'
          - 'benchmarks/django_workload/django-workload/client/*.log'
          - 'benchmarks/django_workload/*.log'
          - '/tmp/wrk_out_*'
          - 'benchmarks/django_workload/breakdown.csv'

- name: feedsim_dlrm
  benchmark: feedsim_autoscale
  description: >
    Single-instance aggregator workload with DLRM (Deep Learning Recommendation
    Model) inference. Same t43 c7 calibration as feedsim_autoscale_dlrm but pinned
    to ONE feedsim instance per host (overrides the autoscale "100 cores per
    instance" rule). Single-instance runs give better gen-over-gen correlation
    than multi-instance — there's no cross-instance mock_services queue contention
    and no per-instance core-range fragmentation. Use this for cross-LSST CPU
    comparisons; use feedsim_autoscale_dlrm for max-throughput experiments on big
    multi-instance boxes. Finds maximum QPS that the system can sustain while
    keeping 95th percentile latency <= 700ms (matches prod multifeed_aggregator's
    own end-to-end p95 budget).
  args:
    - '-n {num_instances}'
    - '--async-io'
    - '--io-dist={io_dist}'
    - '--io-mean={io_mean}'
    - '--workload={workload}'
    - '--dlrm-model={dlrm_model}'
    - '--dlrm-batch-size={dlrm_batch_size}'
    - '--dlrm-threads={dlrm_threads}'
    - '--dlrm-inferences={dlrm_inferences}'
    - '--client-side-features={client_side_features}'
    - '--client-batch-size={client_batch_size}'
    - '--client-inferences={client_inferences}'
    - '--client-feature-seed={client_feature_seed}'
    - '--client-num-dense={client_num_dense}'
    - '--client-num-sparse={client_num_sparse}'
    - '--feature-extractors'
    - '--feature-complexity={feature_complexity}'
    - '--num-stories={num_stories}'
    - '--extractors-per-story={extractors_per_story}'
    - '--story-processors-per-story={story_processors_per_story}'
    - '--stories-per-processor-pass={stories_per_processor_pass}'
    - '--silesia-dir={silesia_dir}'
    - '--stories-per-request={stories_per_request}'
    # t43 c7 calibration (see ~/.claude/projects/.../memory/t43_c7_recommended.md)
    - '--mock-tls={mock_tls}'
    - '--mock-zstd-frac={mock_zstd_frac}'
    - '--mock-keepalive-interval-ms={mock_keepalive_interval_ms}'
    - '--rpc-fanout-scale={rpc_fanout_scale}'
    - '--server-zstd={server_zstd}'
    - '--sla-p95-ms={sla_p95_ms}'
    - '--depth={depth}'
    - '{extra_args}'
  vars:
    # Hardcode num_instances=1: this job exists specifically for single-instance
    # cross-LSST comparisons. If you want autoscale-by-cores, use
    # feedsim_autoscale_dlrm instead.
    - 'num_instances=1'
    - 'io_dist=fixed'
    - 'io_mean=200'
    - 'workload=dlrm'
    - 'dlrm_model=models/dlrm_small.pt'
    - 'dlrm_batch_size=64'
    - 'dlrm_threads=1'
    - 'dlrm_inferences=1'
    - 'client_side_features=0'
    - 'client_batch_size=256'
    - 'client_inferences=64'
    - 'client_feature_seed=42'
    - 'client_num_dense=13'
    - 'client_num_sparse=26'
    - 'feature_complexity=5'
    - 'num_stories=400'
    - 'extractors_per_story=280'
    - 'story_processors_per_story=2'
    - 'stories_per_processor_pass=100'
    - 'silesia_dir=silesia'
    - 'stories_per_request=10'
    - 'mock_tls=1'
    - 'mock_zstd_frac=0.75'
    - 'mock_keepalive_interval_ms=200'
    - 'rpc_fanout_scale=0.05'
    - 'server_zstd=0'
    - 'sla_p95_ms=700'
    # Driver pipeline depth (max outstanding requests per driver connection).
    # Default 1. Raise (e.g. 2) if the final phase saturates neither CPU nor SLA
    # latency (final p95 well below sla_p95_ms while CPU util < ~90%) — often
    # needed on high-perf ARM cores. With adaptive depth on (the default), this is
    # the starting floor the peak search raises from; see README "Driver depth".
    - 'depth=1'
    - 'extra_args='
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/feedsim/feedsim_results*.txt'
          - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
          - 'benchmarks/feedsim/src/perf.data'
          - 'benchmarks/feedsim/breakdown.csv'
          - '/tmp/feedsim_log.txt'

- name: feedsim_autoscale_dlrm
  benchmark: feedsim_autoscale
  description: >
    Multi-instance aggregator workload with DLRM (Deep Learning Recommendation Model)
    inference. Defaults bake in the t43 c7 calibration — the balanced
    configuration that minimizes both gen-over-gen correlation gap AND
    bench-vs-prod hot-function gap on CPL/BGM. Finds maximum QPS that the
    system can sustain while keeping 95th percentile latency <= 700ms
    (matches prod multifeed_aggregator's own end-to-end p95 budget).
    Automatically spawns one workload instance per ~100 cores.
  args:
    - '-n {num_instances}'
    - '--async-io'
    - '--io-dist={io_dist}'
    - '--io-mean={io_mean}'
    - '--workload={workload}'
    - '--dlrm-model={dlrm_model}'
    - '--dlrm-batch-size={dlrm_batch_size}'
    - '--dlrm-threads={dlrm_threads}'
    - '--dlrm-inferences={dlrm_inferences}'
    - '--client-side-features={client_side_features}'
    - '--client-batch-size={client_batch_size}'
    - '--client-inferences={client_inferences}'
    - '--client-feature-seed={client_feature_seed}'
    - '--client-num-dense={client_num_dense}'
    - '--client-num-sparse={client_num_sparse}'
    - '--feature-extractors'
    - '--feature-complexity={feature_complexity}'
    - '--num-stories={num_stories}'
    - '--extractors-per-story={extractors_per_story}'
    - '--story-processors-per-story={story_processors_per_story}'
    - '--stories-per-processor-pass={stories_per_processor_pass}'
    - '--silesia-dir={silesia_dir}'
    - '--stories-per-request={stories_per_request}'
    # t43 c7 calibration knobs (t31 tracking: T269255604)
    - '--mock-tls={mock_tls}'
    - '--mock-zstd-frac={mock_zstd_frac}'
    - '--mock-keepalive-interval-ms={mock_keepalive_interval_ms}'
    - '--rpc-fanout-scale={rpc_fanout_scale}'
    - '--server-zstd={server_zstd}'
    - '--sla-p95-ms={sla_p95_ms}'
    - '--depth={depth}'
    - '{extra_args}'
  vars:
    - 'num_instances=-1'
    - 'io_dist=fixed'
    - 'io_mean=200'
    - 'workload=dlrm'
    - 'dlrm_model=models/dlrm_small.pt'
    # t43 c7: bs=64 is the balanced choice across both gen-over-gen
    # correlation (BGM/CPL = 4.25× → 2nd best of 12 cells) and prod
    # hot-function gap (CPL 4th, BGM 5th). bs=32 wins peak QPS alone but
    # breaks correlation (BGM stretches to 4.75× CPL).
    - 'dlrm_batch_size=64'
    - 'dlrm_threads=1'
    - 'dlrm_inferences=1'
    - 'client_side_features=0'
    - 'client_batch_size=256'
    - 'client_inferences=64'
    - 'client_feature_seed=42'
    - 'client_num_dense=13'
    - 'client_num_sparse=26'
    - 'feature_complexity=5'
    # 400 x 280 = 112K extractor calls/req. Combined with the scaled-up
    # story-processor pipeline (below) this matches prod's FE:story CPU
    # ratio of ~1.3:1 (multifeed: FE 29%, story 22% on BGM Strobelight).
    - 'num_stories=400'
    - 'extractors_per_story=280'
    # Story-processor pipeline scaled 8x from t32 first attempt (was
    # 1 pass/story * 25 stories/pass = 25 cmpns/story → 1.1% Ranking-Story
    # on BGM, vs prod 9.3%). Bumped to 2 passes * 100 stories = 200
    # comparisons/story, target ~9% Ranking-Story share.
    - 'story_processors_per_story=2'
    - 'stories_per_processor_pass=100'
    - 'silesia_dir=silesia'
    - 'stories_per_request=10'
    # t43 c7: TLS+keepalive on mock_services, 75% of channels compress,
    # rpc_fanout_scale=0.05 (~188 RPCs/session). Server-side ZSTD off
    # (mock-side ZSTD provides the prod-realistic Compression share).
    - 'mock_tls=1'
    - 'mock_zstd_frac=0.75'
    - 'mock_keepalive_interval_ms=200'
    - 'rpc_fanout_scale=0.05'
    - 'server_zstd=0'
    - 'sla_p95_ms=700'
    # Driver pipeline depth (max outstanding requests per driver connection).
    # Default 1. Raise (e.g. 2) if the final phase saturates neither CPU nor SLA
    # latency (final p95 well below sla_p95_ms while CPU util < ~90%) — often
    # needed on high-perf ARM cores. With adaptive depth on (the default), this is
    # the starting floor the peak search raises from; see README "Driver depth".
    - 'depth=1'
    - 'extra_args='
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/feedsim/feedsim_results*.txt'
          - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
          - 'benchmarks/feedsim/src/perf.data'
          - 'benchmarks/feedsim/breakdown.csv'
          - '/tmp/feedsim_log.txt'

- name: feedsim_autoscale_dlrm_mini
  benchmark: feedsim_autoscale
  description: >
    DLRM aggregator workload mini version. Configured with a fixed QPS
    for quick validation runs. Uses DLRM inference instead of PageRank
    with async I/O mode.
  args:
    - '-n {num_instances}'
    - '--async-io'
    - '--io-dist={io_dist}'
    - '--io-mean={io_mean}'
    - '--workload={workload}'
    - '--dlrm-model={dlrm_model}'
    - '--dlrm-batch-size={dlrm_batch_size}'
    - '--dlrm-threads={dlrm_threads}'
    - '--dlrm-inferences={dlrm_inferences}'
    - '--client-side-features={client_side_features}'
    - '--client-batch-size={client_batch_size}'
    - '--client-inferences={client_inferences}'
    - '--client-feature-seed={client_feature_seed}'
    - '--client-num-dense={client_num_dense}'
    - '--client-num-sparse={client_num_sparse}'
    - '--feature-extractors'
    - '--feature-complexity={feature_complexity}'
    - '--num-stories={num_stories}'
    - '--extractors-per-story={extractors_per_story}'
    - '-q {fixed_qps}'
    - '-d {fixed_qps_duration}'
    - '-w {warmup_time}'
    - '-r {qps_percentage_threshold}'
    - '-x {max_warmup_iterations}'
    - '-D {queue_drain_time}'
    - '-R {node_rank_seed}'
    - '-P {page_rank_seed}'
    - '-C {pointer_chase_seed}'
    - '-N'
    # t43 c7 calibration knobs (t31 tracking: T269255604)
    - '--mock-tls={mock_tls}'
    - '--mock-zstd-frac={mock_zstd_frac}'
    - '--mock-keepalive-interval-ms={mock_keepalive_interval_ms}'
    - '--rpc-fanout-scale={rpc_fanout_scale}'
    - '--server-zstd={server_zstd}'
    - '--sla-p95-ms={sla_p95_ms}'
    - '--depth={depth}'
    - '{extra_args}'
  vars:
    - 'num_instances=-1'
    - 'io_dist=fixed'
    - 'io_mean=200'
    - 'workload=dlrm'
    - 'dlrm_model=models/dlrm_small.pt'
    - 'dlrm_batch_size=64'
    - 'dlrm_threads=1'
    # 8 inferences/req brings DLRM-Inference CPU share from 55-62% down
    # toward prod multifeed_aggregator's Ranking-Prediction share of 7-13%.
    - 'dlrm_inferences=8'
    - 'client_side_features=0'
    - 'client_batch_size=256'
    - 'client_inferences=64'
    - 'client_feature_seed=42'
    - 'client_num_dense=13'
    - 'client_num_sparse=26'
    - 'feature_complexity=5'
    # 1800 stories x 50 extractors = 90K extractor calls/req. Tuned to
    # push FeatureExtraction CPU share into prod's 30-35% band; 5K
    # calls/req previously gave only 1.7-1.9%.
    - 'num_stories=1800'
    - 'extractors_per_story=50'
    - 'fixed_qps=100000'
    - 'fixed_qps_duration=10'
    - 'warmup_time=5'
    - 'qps_percentage_threshold=-1'
    - 'max_warmup_iterations=-1'
    - 'queue_drain_time=1'
    - 'node_rank_seed=54321'
    - 'page_rank_seed=12345'
    - 'pointer_chase_seed=98765'
    # t43 c7 mock_services + TLS knobs (same baseline as feedsim_autoscale_dlrm)
    - 'mock_tls=1'
    - 'mock_zstd_frac=0.75'
    - 'mock_keepalive_interval_ms=200'
    - 'rpc_fanout_scale=0.05'
    - 'server_zstd=0'
    - 'sla_p95_ms=700'
    # Driver pipeline depth (max outstanding requests per driver connection).
    # Default 1. Raise (e.g. 2) if the final phase saturates neither CPU nor SLA
    # latency (final p95 well below sla_p95_ms while CPU util < ~90%) — often
    # needed on high-perf ARM cores. With adaptive depth on (the default), this is
    # the starting floor the peak search raises from; see README "Driver depth".
    - 'depth=1'
    - 'extra_args='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/feedsim/feedsim_results*.txt'
          - 'benchmarks/feedsim/feedsim-multi-inst-*.log'
          - 'benchmarks/feedsim/src/perf.data'
          - 'benchmarks/feedsim/breakdown.csv'
          - '/tmp/feedsim_log.txt'

- benchmark: spark_standalone
  name: spark_standalone_local
  description: Spark standalone using local SSDs for database and shuffling point
  args:
    - 'run'
    - '--ipv4 {ipv4}'
    - '--real'
  vars:
    - 'ipv4=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/spark_standalone/work'
          - 'benchmarks/spark_standalone/breakdown.csv'

- benchmark: spark_standalone
  name: spark_standalone_remote_setup
  description: Instructions to setup to use remote SSDs via NVMe-over-TCP
  args:
    - 'setup'

- benchmark: spark_standalone
  name: spark_standalone_remote
  description: Spark standalone using remote SSDs for database and shuffling point; compute & memory bound as in prod
  args:
    - 'run'
    - '--dataset-path /flash23/'
    - '--warehouse-dir /flash23/warehouse'
    - '--shuffle-dir /flash23/spark_local_dir'
    - '--ipv4 {ipv4}'
    - '--real'
    - '--sanity {sanity}'
    - '--shuffle-partitions {shuffle_partitions}'
  vars:
    - 'ipv4=0'
    - 'sanity=0'
    - 'shuffle_partitions=200'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/spark_standalone/work'
          - 'benchmarks/spark_standalone/breakdown.csv'

- benchmark: spark_standalone
  name: spark_standalone_remote_mini
  description: Spark standalone using remote SSDs for database and shuffling point; compute & memory bound as in prod
  args:
    - 'run'
    - '--dataset-path /flash23/'
    - '--dataset-name {dataset_name}'
    - '--warehouse-dir /flash23/warehouse'
    - '--shuffle-dir /flash23/spark_local_dir'
    - '--ipv4 {ipv4}'
    - '--real'
    - '--sanity {sanity}'
    - '--local-hostname "{local_hostname}"'
    - '--shuffle-partitions {shuffle_partitions}'
  vars:
    - 'ipv4=0'
    - 'sanity=0'
    - 'dataset_name=bpc_t93586_s2_synthetic_1GB'
    - 'local_hostname='
    - 'shuffle_partitions=20'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/spark_standalone/work'
          - 'benchmarks/spark_standalone/breakdown.csv'

- benchmark: spark_standalone
  name: spark_standalone_remote_compute
  description: Spark standalone using remote SSDs for database and shuffling point; compute bound only
  args:
    - 'run'
    - '--dataset-path /flash23/'
    - '--warehouse-dir /flash23/warehouse'
    - '--shuffle-dir /flash23/spark_local_dir'
    - '--aggressive 1'
    - '--ipv4 {ipv4}'
    - '--local-hostname "{local_hostname}"'
    - '--real'
  vars:
    - 'ipv4=0'
    - 'local_hostname='
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/spark_standalone/work'
          - 'benchmarks/spark_standalone/breakdown.csv'

- benchmark: spark_standalone
  name: spark_standalone_remote_3x
  description: Spark standalone using remote SSDs with 3x query load (2 more joins) and auto-scaled shuffle partitions (3 * CPU cores)
  args:
    - 'run'
    - '--dataset-path /flash23/'
    - '--warehouse-dir /flash23/warehouse'
    - '--shuffle-dir /flash23/spark_local_dir'
    - '--ipv4 {ipv4}'
    - '--real'
    - '--sanity {sanity}'
    - '--shuffle-partitions 0'
    - '--query queries/release_test_93586_3x.sql'
  vars:
    - 'ipv4=0'
    - 'sanity=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/spark_standalone/work'
          - 'benchmarks/spark_standalone/breakdown.csv'


- benchmark: tao_bench
  name: tao_bench_64g
  description:
      Tao benchmark using 64GB memory.
      MAKE SURE to start clients within 1 minute.
  roles:
    server:
      args:
        - 'server'
        - '--memsize=64'
        - '--nic-channel-ratio=0.5'
        - '--fast-threads-ratio=0.75'
        - '--dispatcher-to-fast-ratio=0.25'
        - '--slow-to-fast-ratio=3'
        - '--interface-name={interface_name}'
        - '--real'
      vars:
        - 'interface_name=eth0'
    client:
      args:
        - 'client'
        - '--server-hostname={server_hostname}'
        - '--server-memsize=64'
        - '--target-hit-ratio=0.9'
        - '--tunning-factor=0.807'
        - '--warmup-time=1200'
        - '--test-time=360'
        - '--real'
      vars:
        - 'server_hostname'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'perf.data'

- benchmark: tao_bench
  name: tao_bench_custom
  description:
      Tao benchmark using custom amount of memory and ports.
      MAKE SURE to start clients within 1 minute.
  roles:
    server:
      args:
        - 'server'
        - '--memsize={memsize}'
        - '--nic-channel-ratio=0.5'
        - '--fast-threads-ratio=0.75'
        - '--dispatcher-to-fast-ratio=0.25'
        - '--slow-to-fast-ratio=3'
        - '--interface-name={interface_name}'
        - '--port-number={port_number}'
        - '--warmup-time={warmup_time}'
        - '--test-time={test_time}'
        - '--stats-interval={stats_interval}'
        - '--timeout-buffer={timeout_buffer}'
        - '--disable-tls={disable_tls}'
        - '--smart-nanosleep={smart_nanosleep}'
        - '--memory-file={memory_file}'
        - '--ipv4={ipv4}'
        - '--real'
      vars:
        - 'interface_name=eth0'
        - 'memsize=64'
        - 'port_number=11211'
        - 'warmup_time=1200'
        - 'test_time=360'
        - 'stats_interval=5000'
        - 'timeout_buffer=120'
        - 'disable_tls=0'
        - 'smart_nanosleep=0'
        - 'memory_file='
        - 'ipv4=0'
    client:
      args:
        - 'client'
        - '--clients-per-thread={clients_per_thread}'
        - '--server-hostname={server_hostname}'
        - '--server-memsize={server_memsize}'
        - '--target-hit-ratio=0.9'
        - '--data-size-min={data_size_min}'
        - '--data-size-max={data_size_max}'
        - '--tunning-factor=0.807'
        - '--server-port-number={server_port_number}'
        - '--warmup-time={warmup_time}'
        - '--test-time={test_time}'
        - '--sanity={sanity}'
        - '--wait-after-warmup={wait_after_warmup}'
        - '--disable-tls={disable_tls}'
        - '--ipv4={ipv4}'
        - '--client-id={client_id}'
        - '--control-port={control_port}'
        - '--real'
      vars:
        - 'server_hostname'
        - 'server_memsize=64'
        - 'clients_per_thread=380'
        - 'server_port_number=11211'
        - 'warmup_time=1200'
        - 'test_time=360'
        - 'data_size_min=8191'
        - 'data_size_max=8193'
        - 'wait_after_warmup=5'
        - 'sanity=0'
        - 'disable_tls=0'
        - 'ipv4=0'
        - 'client_id=0'
        - 'control_port=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'perf.data'

- benchmark: tao_bench_autoscale
  name: tao_bench_autoscale
  description:
      Spawns multiple Tao benchmark servers depending on the CPU
      core count. After executing this job, please see the tail
      of benchpress.log for instructions on starting clients.
      MAKE SURE to start clients within 1 minute.
  args:
    - '--num-servers={num_servers}'
    - '--memsize={memsize}'
    - '--fast-threads-ratio={fast_threads_ratio}'
    - '--slow-to-fast-ratio={slow_to_fast_ratio}'
    - '--interface-name={interface_name}'
    - '--affinitize-iolocal-cores={affinitize_iolocal_cores}'
    - '--port-number-start={port_number_start}'
    - '--warmup-time={warmup_time}'
    - '--test-time={test_time}'
    - '--server-hostname={server_hostname}'
    - '--num-clients={num_clients}'
    - '--clients-per-thread={clients_per_thread}'
    - '--bind-cpu={bind_cpu}'
    - '--bind-mem={bind_mem}'
    - '--sanity={sanity}'
    - '--stats-interval={stats_interval}'
    - '--timeout-buffer={timeout_buffer}'
    - '--client-wait-after-warmup={client_wait_after_warmup}'
    - '--disable-tls={disable_tls}'
    - '--smart-nanosleep={smart_nanosleep}'
    - '--memory-file={memory_file}'
    - '--auto-warmup={auto_warmup}'
    - '--ipv4={ipv4}'
    - '--real'
  vars:
    - 'num_servers=0'
    - 'memsize=0'
    - 'port_number_start=11211'
    - 'fast_threads_ratio=0.75'
    - 'slow_to_fast_ratio=3'
    - 'interface_name=eth0'
    - 'affinitize_iolocal_cores=0'
    - 'warmup_time=-1'
    - 'test_time=720'
    - 'server_hostname='
    - 'num_clients=2'
    - 'clients_per_thread=0'
    - 'bind_cpu=1'
    - 'bind_mem=1'
    - 'sanity=0'
    - 'stats_interval=5000'
    - 'timeout_buffer=120'
    - 'client_wait_after_warmup=5'
    - 'disable_tls=0'
    - 'smart_nanosleep=0'
    - 'memory_file='
    - 'auto_warmup=0'
    - 'ipv4=0'
  hooks:
    - hook: tao_instruction
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server*.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'tao-bench-server-*.log'
          - 'perf.data'

- benchmark: tao_bench_standalone
  name: tao_bench_standalone
  description:
      Spawns multiple Tao benchmark servers depending on the CPU
      core count.
      This job starts clients Automatically on the same instance.

  args:
    - '--num-servers={num_servers}'
    - '--memsize={memsize}'
    - '--fast-threads-ratio={fast_threads_ratio}'
    - '--slow-to-fast-ratio={slow_to_fast_ratio}'
    - '--warmup-time={warmup_time}'
    - '--test-time={test_time}'
    - '--num-clients={num_clients}'
    - '--clients-per-thread={clients_per_thread}'
    - '--bind-cpu={bind_cpu}'
    - '--bind-mem={bind_mem}'
    - '--stats-interval={stats_interval}'
    - '--port-number-start={port_number_start}'
    - '--disable-tls={disable_tls}'
    - '--timeout-buffer={timeout_buffer}'
    - '--num-slow-threads={num_slow_threads}'
    - '--num-fast-threads={num_fast_threads}'
    - '--num-client-threads={num_client_threads}'
    - '--memory-file={memory_file}'
    - '--auto-warmup={auto_warmup}'
    - '--auto-fix-ports={auto_fix_ports}'
    - '--auto-fix-ulimit={auto_fix_ulimit}'
    - '--skip-hit-rate-check={skip_hit_rate_check}'
    - '--ipv4={ipv4}'
  vars:
    - 'num_servers=0'
    - 'memsize=0'
    - 'fast_threads_ratio=0.75'
    - 'slow_to_fast_ratio=3'
    - 'warmup_time=-1'
    - 'test_time=720'
    - 'num_clients=2'
    - 'clients_per_thread=0'
    - 'bind_cpu=1'
    - 'bind_mem=1'
    - 'stats_interval=5000'
    - 'port_number_start=11211'
    - 'disable_tls=0'
    - 'timeout_buffer=0'
    - 'num_slow_threads=0'
    - 'num_fast_threads=0'
    - 'num_client_threads=0'
    - 'memory_file='
    - 'auto_warmup=0'
    - 'auto_fix_ports=0'
    - 'auto_fix_ulimit=0'
    - 'skip_hit_rate_check=0'
    - 'ipv4=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server*.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'tao-bench-server-*.log'
          - 'client_*.log'
          - 'perf.data'

- benchmark: tao_bench_standalone
  name: tao_bench_standalone_mini
  description:
      This job starts clients Automatically on the same instance.

  args:
    - '--num-servers={num_servers}'
    - '--memsize={memsize}'
    - '--fast-threads-ratio={fast_threads_ratio}'
    - '--slow-to-fast-ratio={slow_to_fast_ratio}'
    - '--warmup-time={warmup_time}'
    - '--test-time={test_time}'
    - '--num-clients={num_clients}'
    - '--clients-per-thread={clients_per_thread}'
    - '--bind-cpu={bind_cpu}'
    - '--bind-mem={bind_mem}'
    - '--stats-interval={stats_interval}'
    - '--port-number-start={port_number_start}'
    - '--disable-tls={disable_tls}'
    - '--timeout-buffer={timeout_buffer}'
    - '--num-slow-threads={num_slow_threads}'
    - '--num-fast-threads={num_fast_threads}'
    - '--num-client-threads={num_client_threads}'
    - '--warmup-timeout-buffer={warmup_timeout_buffer}'
    - '--test-timeout-buffer={test_timeout_buffer}'
    - '--postprocessing-timeout-buffer={postprocessing_timeout_buffer}'
    - '--poll-interval={poll_interval}'
    - '--memory-file={memory_file}'
    - '--auto-fix-ports={auto_fix_ports}'
    - '--auto-fix-ulimit={auto_fix_ulimit}'
    - '--skip-hit-rate-check={skip_hit_rate_check}'
    - '--ipv4={ipv4}'
  vars:
    - 'num_servers=0'
    - 'memsize=0.5'
    - 'fast_threads_ratio=0.75'
    - 'slow_to_fast_ratio=3'
    - 'warmup_time=0'
    - 'test_time=30'
    - 'num_clients=2'
    - 'clients_per_thread=0'
    - 'bind_cpu=1'
    - 'bind_mem=1'
    - 'stats_interval=5000'
    - 'port_number_start=11211'
    - 'disable_tls=0'
    - 'timeout_buffer=30'
    - 'num_slow_threads=0'
    - 'num_fast_threads=0'
    - 'num_client_threads=0'
    - 'warmup_timeout_buffer=0'
    - 'test_timeout_buffer=0'
    - 'postprocessing_timeout_buffer=60'
    - 'poll_interval=0.2'
    - 'memory_file='
    - 'auto_fix_ports=1'
    - 'auto_fix_ulimit=1'
    - 'skip_hit_rate_check=0'
    - 'ipv4=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server*.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'tao-bench-server-*.log'
          - 'client_*.log'
          - 'perf.data'

- benchmark: tao_bench_autoscale
  name: tao_bench_autoscale_v2_beta
  description:
      Spawns multiple Tao benchmark servers depending on the CPU
      core count. This is an experimental version that aims to
      improve scalability on CPUs with large core counts and reduce
      IRQ overhead.
  args:
    - '--num-servers={num_servers}'
    - '--memsize={memsize}'
    - '--fast-threads-ratio={fast_threads_ratio}'
    - '--slow-to-fast-ratio={slow_to_fast_ratio}'
    - '--slow-threads-use-semaphore={slow_threads_use_semaphore}'
    - '--pin-threads={pin_threads}'
    - '--interface-name={interface_name}'
    - '--port-number-start=11211'
    - '--warmup-time={warmup_time}'
    - '--test-time={test_time}'
    - '--server-hostname={server_hostname}'
    - '--num-clients={num_clients}'
    - '--clients-per-thread={clients_per_thread}'
    - '--client-cores={client_cores}'
    - '--conns-per-server-core={conns_per_server_core}'
    - '--stats-interval={stats_interval}'
    - '--disable-tls={disable_tls}'
    - '--memory-file={memory_file}'
    - '--auto-warmup={auto_warmup}'
    - '--real'
  vars:
    - 'num_servers=0'
    - 'memsize=0'
    - 'fast_threads_ratio=1.25'
    - 'slow_to_fast_ratio=3'
    - 'slow_threads_use_semaphore=1'
    - 'pin_threads=0'
    - 'interface_name=eth0'
    - 'warmup_time=-1'
    - 'test_time=720'
    - 'server_hostname='
    - 'num_clients=2'
    - 'clients_per_thread=0'
    - 'client_cores=0'
    - 'conns_per_server_core=85'
    - 'stats_interval=5000'
    - 'disable_tls=0'
    - 'memory_file='
    - 'auto_warmup=1'
  hooks:
    - hook: tao_instruction
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/tao_bench/server*.csv'
          - 'benchmarks/tao_bench/breakdown.csv'
          - 'tao-bench-server-*.log'
          - 'perf.data'


- name: video_transcode_bench_svt
  benchmark: video_transcode_bench
  description: >
    SVT-AV1 based video encoding workload. Compute and memory intensive.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--max-time {max_time}'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'max_time=0'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_svt_timed
  benchmark: video_transcode_bench
  description: >
    Time-based SVT-AV1 video encoding workload. Runs encoding for a fixed
    duration rather than a fixed amount of work, ensuring core saturation
    regardless of core count.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--sample-rate {sample_rate}'
    - '--sampling-seed {sampling_seed}'
    - '--max-time {max_time}'
    - '--score-mode {score_mode}'
    - '--fast-jobs-first {fast_jobs_first}'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'sample_rate=1.0'
    - 'sampling_seed=1000'
    - 'max_time=600'
    - 'score_mode=megapixel'
    - 'fast_jobs_first=1'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_svt_timed_prep
  benchmark: video_transcode_bench
  description: >
    Prep run for time-based SVT-AV1 video encoding workload. Downscales all
    clips and caches them in resized_clips/ for subsequent runs with
    video_transcode_bench_svt_timed_reuse. After this run, the datasets/cuts/
    directory (~42GB) can be deleted to save space.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--sample-rate {sample_rate}'
    - '--sampling-seed {sampling_seed}'
    - '--max-time {max_time}'
    - '--score-mode {score_mode}'
    - '--fast-jobs-first {fast_jobs_first}'
    - '--keep-downscaled'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'sample_rate=1.0'
    - 'sampling_seed=1000'
    - 'max_time=600'
    - 'score_mode=megapixel'
    - 'fast_jobs_first=0'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'
          - '1'
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_svt_timed_reuse
  benchmark: video_transcode_bench
  description: >
    Reuse run for time-based SVT-AV1 video encoding workload. Skips
    downscaling and uses cached resized_clips/ from a prior prep run
    (video_transcode_bench_svt_timed_prep). Errors if resized_clips/
    is not found. Keeps cached clips for subsequent runs.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--max-time {max_time}'
    - '--score-mode {score_mode}'
    - '--fast-jobs-first {fast_jobs_first}'
    - '--skip-downscale'
    - '--keep-downscaled'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'max_time=600'
    - 'score_mode=megapixel'
    - 'fast_jobs_first=0'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'
          - '1'
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_svt_mini
  benchmark: video_transcode_bench
  description: >
    SVT-AV1 based video encoding workload. Compute and memory intensive.
  execute_cmd_from_file: true
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--sample-rate {sample_rate}'
    - '--sampling-seed {sampling_seed}'
    - '--sleep-before-perf {sleep_before_perf}'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'sample_rate=0.01'
    - 'sampling_seed=1000'
    - 'sleep_before_perf=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'


- name: video_transcode_bench_svt_timed_mini_prep
  benchmark: video_transcode_bench
  description: >
    Prep run for time-based mini SVT-AV1 video encoding workload. Downscales
    a sampled subset of clips (20%) and caches them in resized_clips/ for
    subsequent runs with video_transcode_bench_svt_timed_mini_reuse. After
    this run, the datasets/cuts/ directory (~42GB) can be deleted to save space.
  execute_cmd_from_file: true
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--sample-rate {sample_rate}'
    - '--sampling-seed {sampling_seed}'
    - '--sleep-before-perf {sleep_before_perf}'
    - '--max-time {max_time}'
    - '--score-mode {score_mode}'
    - '--fast-jobs-first {fast_jobs_first}'
    - '--keep-downscaled'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'sample_rate=0.2'
    - 'sampling_seed=1000'
    - 'sleep_before_perf=0'
    - 'max_time=600'
    - 'score_mode=megapixel'
    - 'fast_jobs_first=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_svt_timed_mini_reuse
  benchmark: video_transcode_bench
  description: >
    Reuse run for time-based mini SVT-AV1 video encoding workload. Skips
    downscaling and uses cached resized_clips/ from a prior prep run
    (video_transcode_bench_svt_timed_mini_prep). Errors if resized_clips/
    is not found. Keeps cached clips for subsequent runs.
  execute_cmd_from_file: true
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
    - '--parallelism {parallelism}'
    - '--procs {procs}'
    - '--sleep-before-perf {sleep_before_perf}'
    - '--max-time {max_time}'
    - '--score-mode {score_mode}'
    - '--fast-jobs-first {fast_jobs_first}'
    - '--skip-downscale'
    - '--keep-downscaled'
  vars:
    - 'encoder=svt'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
    - 'parallelism=1'
    - 'procs=-1'
    - 'sleep_before_perf=0'
    - 'max_time=8'
    - 'score_mode=megapixel'
    - 'fast_jobs_first=0'
  hooks:
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_aom
  benchmark: video_transcode_bench
  description: >
    AOM based video encoding workload. Compute and memory intensive.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
  vars:
    - 'encoder=aom'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- name: video_transcode_bench_x264
  benchmark: video_transcode_bench
  description: >
    x264 based video encoding workload. Compute and memory intensive.
  args:
    - '--encoder {encoder}'
    - '--levels {levels}'
    - '--output {output}'
    - '--runtime {runtime}'
  vars:
    - 'encoder=x264'
    - 'levels=0:0'
    - 'output=video_transcode_bench_results.txt'
    - 'runtime=medium'
  hooks:
    - hook: cpu-mpstat
      options:
        args:
          - '-u'   # utilization
          - '1'    # second interval
    - hook: copymove
      options:
        is_move: true
        after:
          - 'benchmarks/video_transcode_bench/video_transcode_bench_results.txt'
          - 'benchmarks/video_transcode_bench/perf.data'
          - 'benchmarks/video_transcode_bench/breakdown.csv'

- benchmark: health_check
  name: health_check
  description: Default run for health_check
  roles:
    server:
      args:
        - -r server
        - '-c {clients}'
      vars:
        - 'clients'
    client:
      args:
        - '-r client'

- benchmark: health_check
  name: mm_mem
  description: Standalone run for mm-mem memory latency and bandwidth benchmark
  args:
    - '-b mm-mem'
    - '--'
    - '{extra_args}'
  vars:
    - 'extra_args='

- benchmark: health_check
  name: mm_mem_peak_bw
  description: Run mm-mem peak bandwidth and memcpy benchmark
  args:
    - '-b mm-mem'
    - '--'
    - '--test bandwidth'
    - '{extra_args}'
  vars:
    - 'extra_args='

- benchmark: health_check
  name: mm_mem_idle_latency
  description: Run mm-mem idle latency benchmark
  args:
    - '-b mm-mem'
    - '--'
    - '--test idle_latency'
    - '{extra_args}'
  vars:
    - 'extra_args='

- benchmark: health_check
  name: mm_mem_loaded_latency
  description: Run mm-mem loaded latency benchmark
  args:
    - '-b mm-mem'
    - '--'
    - '--test loaded_latency'
    - '{extra_args}'
  vars:
    - 'extra_args='

- benchmark: sleep
  name: sleep
  description: >
    Sleep for a period of time. Can be used to leverage Benchpress's Perf
    Monitoring Hook to monitor some external workloads.
  args:
    - '{duration}'
  vars:
    - 'duration'

- benchmark: graph500_omp_csr
  name: graph500_omp_csr
  description: >
    Graph500 v3 reference BFS (MPI). Single positional arg is SCALE (log2
    vertices); run.sh wraps the binary with mpirun -np <largest pow2 <= nproc>.
    SCALE defaults to 20 and can be overridden per-run from the CLI, e.g.
    `./benchpress run graph500_omp_csr --role_input='{"scale":"25"}'`.
  args:
    - "{scale}"
  vars:
    - "scale=20"

- benchmark: liblinear
  name: liblinear_synthetic
  description: >
    liblinear training + prediction on a synthetic arbitrary-size LIBSVM
    dataset. Size/shape are overridable, e.g.
    `./benchpress run liblinear_synthetic -i '{"samples": "500000"}'`.
  args:
    dataset: synthetic
    samples: "{samples}"
    features: "{features}"
    density: "{density}"
    noise: "{noise}"
    seed: "{seed}"
    solver: "{solver}"
    cost: "{cost}"
    bias: "{bias}"
  vars:
    - 'samples=200000'
    - 'features=1000'
    - 'density=0.1'
    - 'noise=0.0'
    - 'seed=42'
    - 'solver=1'
    - 'cost=1'
    - 'bias=1'
  hooks: []

- benchmark: liblinear
  name: liblinear_rcv1
  description: liblinear training + prediction on the real rcv1.binary dataset
  args:
    dataset: rcv1
    solver: 1
    cost: 1
  hooks: []