wb | profiles: new utxoscale-solo profile family

IntersectMBO · May 8, 2024 · a26fb90 · a26fb90
1 parent f255977
commit a26fb90
Show file tree

Hide file tree

Showing 3 changed files with 75 additions and 3 deletions.
diff --git a/Makefile b/Makefile
@@ -101,6 +101,8 @@ PROFILES_NOMAD_PERF       += plutus-nomadperf fast-nomadperf latency-nomadperf
 PROFILES_NOMAD_PERF_NOP2P := default-nomadperf-nop2p oldtracing-nomadperf-nop2p ci-test-nomadperf-nop2p ci-bench-nomadperf-nop2p
 PROFILES_NOMAD_PERF_NOP2P += value-nomadperf-nop2p value-oldtracing-nomadperf-nop2p plutus-nomadperf-nop2p fast-nomadperf-nop2p
 PROFILES_NOMAD_PERFSSD    := fast-nomadperfssd
+# single node profiles on the NomadSSD cluster on AWS
+PROFILES_UTXOSCALE_SOLO	  := utxoscale-solo-24M64G-nomadperfssd utxoscale-solo-12M64G-nomadperfssd utxoscale-solo-12M16G-nomadperfssd
 
 LOCAL_PROFILES += $(PROFILES_BASE)
 LOCAL_PROFILES += $(PROFILES_FAST)
@@ -120,6 +122,7 @@ LOCAL_PROFILES += $(PROFILES_VENDOR)
 CLOUD_PROFILES += $(PROFILES_NOMAD_PERF)
 CLOUD_PROFILES += $(PROFILES_NOMAD_PERF_NOP2P)
 CLOUD_PROFILES += $(PROFILES_NOMAD_PERFSSD)
+CLOUD_PROFILES += $(PROFILES_UTXOSCALE_SOLO)
 
 
 ## Note:  to enable a shell for a profile, just add its name (one of names from 'make ps') to SHELL_PROFILES

diff --git a/nix/workbench/profile/prof1-variants.jq b/nix/workbench/profile/prof1-variants.jq
@@ -1,5 +1,18 @@
 import "epoch-timeline" as timeline;
 
+## For the Nomad perf-ssd cluster, we might want to artificially
+## cap the large RAM resources the instances provide.
+def nomad_memory_limit($limit):
+  { nomad:
+    { resources:
+      { producer:
+        { memory:       $limit
+        , memory_max:   $limit
+        }
+      }
+    }
+  };
+
 def all_profile_variants:
                                          1024    as $Ki
   |                                      1000000 as $M
@@ -281,8 +294,8 @@ def all_profile_variants:
     } as $compressed_timescale
   |
     { genesis:
-      { epoch_length:                   1800
-      , parameter_k:                    9
+      { epoch_length:                   1200
+      , parameter_k:                    6
       }
     } as $small_timescale
   |
@@ -547,6 +560,10 @@ def all_profile_variants:
    ($model_timescale * $nomad_perf_tps_saturation_value *
     { scenario:                        "fixed-loaded"
     }) as $scenario_nomad_perf
+  |
+   ($small_timescale * $nomad_perf_tps_saturation_value *
+    { scenario:                        "fixed-loaded"
+    }) as $scenario_nomad_perfssd_solo
   |
    ($model_timescale * $model_tps_saturation_value *
     { scenario:                        "fixed-loaded"
@@ -636,6 +653,23 @@ def all_profile_variants:
    ($scenario_latency * $compose_fiftytwo * $dataset_empty * $no_filtering *
     { desc: "AWS c5-2xlarge cluster, stop when all latency services stop"
     }) as $nomad_perf_latency_base
+  |
+   ($scenario_nomad_perfssd_solo * $solo * $dataset_24m *
+    { node:
+        { shutdown_on_slot_synced:        7200
+        }
+      , analysis:
+        { filters:                        ["epoch3+", "size-full"]
+        }
+      , generator:
+        { epochs:                         6
+        }
+      , genesis:
+        { funds_balance:                  20000000000000
+        , max_block_size:                 88000
+        }
+      , desc: "AWS c5[d]-9xlarge utxoscale dataset, 6 epochs"
+    }) as $nomad_perfssd_solo_base
   |
    ($scenario_model * $quadruplet * $dataset_current * $for_7ep *
     { node:
@@ -715,6 +749,13 @@ def all_profile_variants:
   ### Actual profiles
   ##
 
+  ### Profile templates
+  ###
+  # UTxO scaling on a single node, mainnet blocksize, ~2h runtime (6 epochs) - default: 24mio UTxO, 64GB RAM cap
+    ($nomad_perfssd_solo_base * $nomad_perfssd_unicircle * $costmodel_v8_preview * $p2p
+    ) as $utxoscale_solo_template
+  |
+
   ### First, auto-named profiles:
   ###
   ## Short slots:
@@ -970,6 +1011,24 @@ def all_profile_variants:
     { name: "fast-nomadperfssd"
     }
 
+## P&T NomadSSD cluster: UTxO scale benchmarks on a single node
+  , $utxoscale_solo_template *
+    { name: "utxoscale-solo-24M64G-nomadperfssd"
+    }
+  , $utxoscale_solo_template *
+    { name: "utxoscale-solo-12M64G-nomadperfssd"
+    , genesis:
+      { utxo:                               (12 * $M)
+      }
+    }
+  , $utxoscale_solo_template *
+    { name: "utxoscale-solo-12M16G-nomadperfssd"
+    , genesis:
+      { utxo:                               (12 * $M)
+      }
+    , cluster:                              nomad_memory_limit(16384)
+    }
+
 ## Model value variant: 7 epochs (128GB RAM needed; 16GB for testing locally)
   , $model_base * $costmodel_v8_preview *
     { name: "model-value"

diff --git a/nix/workbench/profile/prof3-derived.jq b/nix/workbench/profile/prof3-derived.jq
@@ -85,6 +85,16 @@ def add_derived_params:
 | .era                                       as $era
 | .node                                      as $node
 | .genesis.shelley.protocolParams            as $pparams
+| .cluster.nomad                             as $nomad
+
+## The perf-ssd machines have abundant physical RAM, and Nomad uses cgroups to constrain resources.
+## To also influence RTS / GC behaviour, -M needs to be used, as the RTS infers a heap limit from
+## the system's ulimit, not the cgroup limit.
+| $node.rts_flags_override                   as $rtsflags
+| (if $nomad.class == "perf-ssd"
+   then $rtsflags + [("-M" + ($nomad.resources.producer.memory_max | tostring) + "M")]
+   else $rtsflags
+   end)                                      as $rtsflags_derived
 
 ## Absolute durations:
 | ($gsis.epoch_length * $gsis.slot_duration)       as $epoch_duration
@@ -191,7 +201,7 @@ def add_derived_params:
          { tx_count:              $generator_tx_count
          }
      , node:
-         {
+         { rts_flags_override:    $rtsflags_derived
          }
      , analysis:
          { minimum_chain_density:      ($gsis.active_slots_coeff * 0.5)