erlang: ring benchmark + results — Phase 3 closed

lib/erlang/bench_ring.sh

@@ -0,0 +1,86 @@
+#!/usr/bin/env bash
+# Erlang-on-SX ring benchmark.
+#
+# Spawns N processes in a ring, passes a token N hops (one full round),
+# and reports wall-clock time + throughput. Aspirational target from
+# the plan is 1M processes; current sync-scheduler architecture caps out
+# orders of magnitude lower — this script measures honestly across a
+# range of N so the result/scaling is recorded.
+#
+# Usage:
+#   bash lib/erlang/bench_ring.sh                # default ladder
+#   bash lib/erlang/bench_ring.sh 100 1000 5000  # custom Ns
+
+set -uo pipefail
+cd "$(git rev-parse --show-toplevel)"
+
+SX_SERVER="${SX_SERVER:-hosts/ocaml/_build/default/bin/sx_server.exe}"
+if [ ! -x "$SX_SERVER" ]; then
+  SX_SERVER="/root/rose-ash/hosts/ocaml/_build/default/bin/sx_server.exe"
+fi
+if [ ! -x "$SX_SERVER" ]; then
+  echo "ERROR: sx_server.exe not found." >&2
+  exit 1
+fi
+
+if [ "$#" -gt 0 ]; then
+  NS=("$@")
+else
+  NS=(10 100 500 1000)
+fi
+
+TMPFILE=$(mktemp)
+trap "rm -f $TMPFILE" EXIT
+
+# One-line Erlang program. Replaces __N__ with the size for each run.
+PROGRAM='Me = self(), N = __N__, Spawner = fun () -> receive {setup, Next} -> Loop = fun () -> receive {token, 0, Parent} -> Parent ! done; {token, K, Parent} -> Next ! {token, K-1, Parent}, Loop() end end, Loop() end end, BuildRing = fun (K, Acc) -> if K =:= 0 -> Acc; true -> BuildRing(K-1, [spawn(Spawner) | Acc]) end end, Pids = BuildRing(N, []), Wire = fun (Ps) -> case Ps of [P, Q | _] -> P ! {setup, Q}, Wire(tl(Ps)); [Last] -> Last ! {setup, hd(Pids)} end end, Wire(Pids), hd(Pids) ! {token, N, Me}, receive done -> done end'
+
+run_n() {
+  local n="$1"
+  local prog="${PROGRAM//__N__/$n}"
+  cat > "$TMPFILE" <<EPOCHS
+(epoch 1)
+(load "lib/erlang/tokenizer.sx")
+(load "lib/erlang/parser.sx")
+(load "lib/erlang/parser-core.sx")
+(load "lib/erlang/parser-expr.sx")
+(load "lib/erlang/parser-module.sx")
+(load "lib/erlang/transpile.sx")
+(load "lib/erlang/runtime.sx")
+(epoch 2)
+(eval "(erlang-eval-ast \"${prog//\"/\\\"}\")")
+EPOCHS
+
+  local start_s start_ns end_s end_ns elapsed_ms
+  start_s=$(date +%s)
+  start_ns=$(date +%N)
+  out=$(timeout 300 "$SX_SERVER" < "$TMPFILE" 2>&1)
+  end_s=$(date +%s)
+  end_ns=$(date +%N)
+
+  local ok="false"
+  if echo "$out" | grep -q ':name "done"'; then ok="true"; fi
+
+  # ms = (end_s - start_s)*1000 + (end_ns - start_ns)/1e6
+  elapsed_ms=$(awk -v s1="$start_s" -v n1="$start_ns" -v s2="$end_s" -v n2="$end_ns" \
+    'BEGIN { printf "%d", (s2 - s1) * 1000 + (n2 - n1) / 1000000 }')
+
+  if [ "$ok" = "true" ]; then
+    local hops_per_s
+    hops_per_s=$(awk -v n="$n" -v ms="$elapsed_ms" \
+      'BEGIN { if (ms == 0) ms = 1; printf "%.0f", n * 1000 / ms }')
+    printf "  N=%-8s  hops=%-8s  %sms  (%s hops/s)\n" "$n" "$n" "$elapsed_ms" "$hops_per_s"
+  else
+    printf "  N=%-8s  FAILED  %sms\n" "$n" "$elapsed_ms"
+  fi
+}
+
+echo "Ring benchmark — sx_server.exe (synchronous scheduler)"
+echo
+for n in "${NS[@]}"; do
+  run_n "$n"
+done
+echo
+echo "Note: 1M-process target from the plan is aspirational; the synchronous"
+echo "scheduler with shift-based suspension and dict-based env copies is not"
+echo "engineered for that scale. Numbers above are honest baselines."

lib/erlang/bench_ring_results.md

@@ -0,0 +1,35 @@
+# Ring Benchmark Results
+
+Generated by `lib/erlang/bench_ring.sh` against `sx_server.exe` on the
+synchronous Erlang-on-SX scheduler.
+
+| N (processes) | Hops | Wall-clock | Throughput |
+|---|---|---|---|
+| 10 | 10 | 907ms | 11 hops/s |
+| 50 | 50 | 2107ms | 24 hops/s |
+| 100 | 100 | 3827ms | 26 hops/s |
+| 500 | 500 | 17004ms | 29 hops/s |
+| 1000 | 1000 | 29832ms | 34 hops/s |
+
+(Each `Nm` row spawns N processes connected in a ring and passes a
+single token N hops total — i.e. the token completes one full lap.)
+
+## Status of the 1M-process target
+
+Phase 3's stretch goal in `plans/erlang-on-sx.md` is a million-process
+ring benchmark. **That target is not met** in the current synchronous
+scheduler; extrapolating from the table above, 1M hops would take
+~30 000 s. Correctness is fine — the program runs at every measured
+size — but throughput is bound by per-hop overhead.
+
+Per-hop cost is dominated by:
+- `er-env-copy` per fun clause attempt (whole-dict copy each time)
+- `call/cc` capture + `raise`/`guard` unwind on every `receive`
+- `er-q-delete-at!` rebuilds the mailbox backing list on every match
+- `dict-set!`/`dict-has?` lookups in the global processes table
+
+To reach 1M-process throughput in this architecture would need at
+least: persistent (path-copying) envs, an inline scheduler that
+doesn't call/cc on the common path (msg-already-in-mailbox), and a
+linked-list mailbox. None of those are in scope for the Phase 3
+checkbox — captured here as the floor we're starting from.