Priority Queue Benchmarks

Summary:
Refactored the existing benchmark to measure more of what is actually happening:

1. Insert writable streams
2. Get the head writable stream for a while
3. Erase the head stream

Addeed a few benchmarks for insert/erase and different queue sizes

Reviewed By: mjoras

Differential Revision: D69100615

fbshipit-source-id: 209a4d1eccf15b482fd1880a3f09706f1361c53b

quic/priority/test/BUCK

@@ -1,4 +1,4 @@
-load("@fbcode//quic:defs.bzl", "mvfst_cpp_test")
+load("@fbcode//quic:defs.bzl", "mvfst_cpp_benchmark", "mvfst_cpp_test")
 
 # Common library
 
@@ -36,3 +36,13 @@ mvfst_cpp_test(
         "//quic/priority:http_priority_queue",
     ],
 )
+
+mvfst_cpp_benchmark(
+    name = "priority_queue_benchmark",
+    srcs = ["QuicPriorityQueueBenchmark.cpp"],
+    deps = [
+        "//common/init:init",
+        "//folly:benchmark",
+        "//quic/priority:http_priority_queue",
+    ],
+)

quic/priority/test/QuicPriorityQueueBenchmark.cpp

@@ -0,0 +1,183 @@
+/*
+ * 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.
+ */
+
+#include <common/init/Init.h>
+#include <folly/Benchmark.h>
+#include <quic/priority/HTTPPriorityQueue.h>
+#include <vector>
+
+using namespace std;
+using namespace folly;
+
+static inline void insert(
+    quic::HTTPPriorityQueue& pq,
+    size_t numConcurrentStreams,
+    bool incremental) {
+  // insert streams at various priorities
+  for (size_t i = 0; i < numConcurrentStreams; i++) {
+    pq.insertOrUpdate(
+        quic::PriorityQueue::Identifier::fromStreamID(i),
+        quic::HTTPPriorityQueue::Priority(i % 8, incremental));
+  }
+}
+
+static inline void processQueueIncremental(
+    quic::HTTPPriorityQueue& pq,
+    size_t numConcurrentStreams,
+    size_t packetsPerStream,
+    uint8_t shift) {
+  CHECK_GT(packetsPerStream, 0);
+  CHECK_EQ(numConcurrentStreams % 8, 0) << "requires equal streams per urgency";
+
+  for (uint8_t urgency = 0; urgency < 8; urgency++) {
+    for (size_t i = 0;
+         i < (numConcurrentStreams / 8 + shift) * (packetsPerStream - 1);
+         i++) {
+      (void)pq.getNextScheduledID(quic::none);
+    }
+    for (size_t i = 0; i < (numConcurrentStreams / 8); i++) {
+      auto id = pq.getNextScheduledID(quic::none);
+      // LOG(INFO) << id.asStreamID();
+      pq.erase(id);
+    }
+  }
+}
+
+static inline void processQueueSequential(
+    quic::HTTPPriorityQueue& pq,
+    size_t numConcurrentStreams,
+    size_t packetsPerStream) {
+  CHECK_GT(packetsPerStream, 0);
+  for (size_t i = 0; i < numConcurrentStreams; i++) {
+    quic::PriorityQueue::Identifier id;
+    for (size_t p = 0; p < packetsPerStream; p++) {
+      id = pq.getNextScheduledID(quic::none);
+      // LOG(INFO) << id.asStreamID();
+    }
+    pq.erase(id);
+  }
+}
+
+static inline void benchmarkPriority(
+    size_t numConcurrentStreams,
+    bool incremental) {
+  quic::HTTPPriorityQueue pq;
+  insert(pq, numConcurrentStreams, incremental);
+
+  size_t packetsPerStream = 4;
+  if (incremental) {
+    processQueueIncremental(pq, numConcurrentStreams, packetsPerStream, 1);
+  } else {
+    processQueueSequential(pq, numConcurrentStreams, packetsPerStream);
+  }
+  CHECK(pq.empty());
+}
+
+BENCHMARK(sequential, n) {
+  for (size_t i = 0; i < n; i++) {
+    benchmarkPriority(96, false);
+  }
+}
+
+BENCHMARK(sequentialCrossover, n) {
+  for (size_t i = 0; i < n; i++) {
+    benchmarkPriority(40, false);
+  }
+}
+
+BENCHMARK(incremental, n) {
+  for (size_t i = 0; i < n; i++) {
+    benchmarkPriority(96, true);
+  }
+}
+BENCHMARK(sequential8, n) {
+  for (size_t i = 0; i < n; i++) {
+    benchmarkPriority(8, false);
+  }
+}
+
+BENCHMARK(incremental8, n) {
+  for (size_t i = 0; i < n; i++) {
+    benchmarkPriority(8, true);
+  }
+}
+
+BENCHMARK(insertSequential, n) {
+  // insert streams at various priorities
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    insert(pq, 100, false);
+    pq.clear();
+  }
+}
+
+BENCHMARK(insertIncremental, n) {
+  // insert streams at various priorities
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    insert(pq, 100, true);
+    pq.clear();
+  }
+}
+
+BENCHMARK(processSequential, n) {
+  // insert streams at various priorities
+  size_t nStreams = 96;
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    BENCHMARK_SUSPEND {
+      insert(pq, nStreams, false);
+    }
+    processQueueSequential(pq, nStreams, 4);
+  }
+}
+
+BENCHMARK(processIncremental, n) {
+  // insert streams at various priorities
+  size_t nStreams = 96;
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    BENCHMARK_SUSPEND {
+      insert(pq, nStreams, true);
+    }
+    processQueueIncremental(pq, nStreams, 4, 0);
+  }
+}
+
+BENCHMARK(eraseSequential, n) {
+  // insert streams at various priorities
+  size_t nStreams = 96;
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    BENCHMARK_SUSPEND {
+      insert(pq, nStreams, false);
+    }
+    while (!pq.empty()) {
+      pq.erase(pq.getNextScheduledID(quic::none));
+    }
+  }
+}
+
+BENCHMARK(eraseIncremental, n) {
+  // insert streams at various priorities
+  size_t nStreams = 96;
+  for (size_t j = 0; j < n; j++) {
+    quic::HTTPPriorityQueue pq;
+    BENCHMARK_SUSPEND {
+      insert(pq, nStreams, true);
+    }
+    while (!pq.empty()) {
+      pq.erase(pq.getNextScheduledID(quic::none));
+    }
+  }
+}
+
+int main(int argc, char** argv) {
+  facebook::initFacebook(&argc, &argv);
+  runBenchmarks();
+  return 0;
+}