From cd961cc231ca3073f03d82b99153a6d0f5a4c7f1 Mon Sep 17 00:00:00 2001
From: Douglas Rumbaugh <dbr4@psu.edu>
Date: Sun, 25 May 2025 14:15:36 -0400
Subject: Added selectivity sweep benchmark

---
 benchmarks/include/file_util.h                |  20 ++++
 benchmarks/tail-latency/selectivity_sweep.cpp | 134 ++++++++++++++++++++++++++
 2 files changed, 154 insertions(+)
 create mode 100644 benchmarks/tail-latency/selectivity_sweep.cpp

(limited to 'benchmarks')

diff --git a/benchmarks/include/file_util.h b/benchmarks/include/file_util.h
index 39b297b..2d340e6 100644
--- a/benchmarks/include/file_util.h
+++ b/benchmarks/include/file_util.h
@@ -86,6 +86,26 @@ static std::vector<QP> read_range_queries(std::string &fname,
   return queries;
 }
 
+template <typename QP>
+static std::vector<QP> generate_uniform_range_queries(size_t n, size_t key_limit, double selectivity) {
+  std::vector<QP> queries;
+
+  gsl_rng *rng = gsl_rng_alloc(gsl_rng_mt19937);
+  size_t query_size = key_limit * selectivity;
+
+  for (size_t i=0; i<n; i++) {
+    QP q;
+    q.lower_bound = gsl_rng_uniform_int(rng, key_limit - query_size);
+    q.upper_bound = q.lower_bound + query_size;
+
+    queries.push_back(q);
+  }
+
+  gsl_rng_free(rng);
+
+  return queries;
+}
+
 template <typename QP>
 static std::vector<QP> read_sosd_point_lookups(std::string &fname, size_t n) {
   std::vector<QP> queries;
diff --git a/benchmarks/tail-latency/selectivity_sweep.cpp b/benchmarks/tail-latency/selectivity_sweep.cpp
new file mode 100644
index 0000000..4faf694
--- /dev/null
+++ b/benchmarks/tail-latency/selectivity_sweep.cpp
@@ -0,0 +1,134 @@
+/*
+ *
+ */
+
+#include "framework/scheduling/SerialScheduler.h"
+#include "framework/util/Configuration.h"
+#include "util/types.h"
+#define ENABLE_TIMER
+#define TS_TEST
+
+#include <thread>
+
+#include "framework/DynamicExtension.h"
+#include "framework/scheduling/FIFOScheduler.h"
+#include "shard/ISAMTree.h"
+#include "query/rangecount.h"
+#include "framework/interface/Record.h"
+#include "file_util.h"
+#include "standard_benchmarks.h"
+
+#include "framework/reconstruction/FixedShardCountPolicy.h"
+
+#include <gsl/gsl_rng.h>
+
+#include "psu-util/timer.h"
+
+
+typedef de::Record<uint64_t, uint64_t> Rec;
+typedef de::ISAMTree<Rec> Shard;
+typedef de::rc::Query<Shard> Q;
+typedef de::DynamicExtension<Shard, Q, de::DeletePolicy::TOMBSTONE, de::SerialScheduler> Ext;
+typedef Q::Parameters QP;
+typedef de::DEConfiguration<Shard, Q, de::DeletePolicy::TOMBSTONE, de::SerialScheduler> Conf;
+
+void usage(char *progname) {
+    fprintf(stderr, "%s reccnt\n", progname);
+}
+
+int main(int argc, char **argv) {
+
+    if (argc < 2) {
+        usage(argv[0]);
+        exit(EXIT_FAILURE);
+    }
+
+    size_t n = atol(argv[1]);
+    std::string d_fname = "unif";
+    
+    auto data = read_sosd_file<Rec>(d_fname, n);
+
+    std::vector<std::vector<QP>> query_sets;
+    std::vector<double> selectivities = {0.0000001, 0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, .25};
+
+    for (auto sel: selectivities) {
+        query_sets.push_back(generate_uniform_range_queries<QP>(100, n, sel));
+    }
+
+    std::vector<size_t> sfs = {2, 4, 8}; //, 4, 8, 16, 32, 64, 128, 256, 512, 1024}; 
+    size_t buffer_size = 8000;
+    std::vector<size_t> policies = {0, 1};
+
+    for (auto pol: policies) {
+    for (size_t i=0; i<sfs.size(); i++) {
+        auto policy = get_policy<Shard, Q>(sfs[i], buffer_size, pol, n);
+        auto config = Conf(std::move(policy));
+        config.recon_enable_maint_on_flush = false;
+        config.recon_maint_disabled = true;
+        config.buffer_flush_trigger = 4000;
+        config.maximum_threads = 8;
+        
+        auto extension = new Ext(std::move(config));
+
+        /* warmup structure w/ 10% of records */
+        size_t warmup = .1 * n;
+        for (size_t j=0; j<warmup; j++) {
+            while (!extension->insert(data[j])) {
+                usleep(1);
+            }
+        }
+
+        extension->await_version();
+
+        // fprintf(stderr, "\n[I] Running Insertion Benchmark\n");
+
+        TIMER_INIT();
+
+        TIMER_START();
+        for (size_t j=warmup; j<data.size(); j++) {
+            while (!extension->insert(data[j])) {
+                usleep(1);
+                fprintf(stderr, "insert blocked %ld\r", j);
+            }
+        }
+        TIMER_STOP();
+        auto total_insert_lat = TIMER_RESULT();
+
+        // extension->print_structure();
+        // fflush(stdout);
+
+        // fprintf(stderr, "\n[I] Finished running insertion benchmark\n");
+        extension->await_version();
+
+        // fprintf(stderr, "[I] Running query benchmark\n");
+        
+        /* repeat the queries a bunch of times */
+
+        auto insert_throughput =  (size_t) ((double) (n - warmup) / (double) total_insert_lat *1.0e9);
+        fprintf(stdout, "%ld\t%ld\t%ld\t%ld\t", pol, sfs[i], n, insert_throughput);
+
+        size_t total = 0;
+        for (size_t l=0; l<query_sets.size(); l++) {
+            TIMER_START();
+            for (size_t f=0; f<query_sets[l].size()*10; f++) {
+                auto q = query_sets[l][f%10];
+                auto res = extension->query(std::move(q));
+                total += res.get();
+            }
+            TIMER_STOP();
+            auto query_latency = (TIMER_RESULT()) / (10*query_sets[l].size());
+            fprintf(stdout, "%lf\t%ld\t", selectivities[l], query_latency);
+        }
+
+        fprintf(stdout, "\n");
+        fprintf(stderr, "%ld\n", total);
+        fflush(stdout);
+
+        // extension->print_structure();
+        delete extension;
+    }
+    }
+
+    fflush(stderr);
+}
+
-- 
cgit v1.2.3