From f24fdf2fd310a5f868e15cd9682ca37d740c77af Mon Sep 17 00:00:00 2001
From: Douglas Rumbaugh <dbr4@psu.edu>
Date: Tue, 30 Jan 2024 15:31:03 -0500
Subject: Benchmarking updates

---
 benchmarks/insert_tail_latency.cpp | 80 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 80 insertions(+)
 create mode 100644 benchmarks/insert_tail_latency.cpp

(limited to 'benchmarks/insert_tail_latency.cpp')

diff --git a/benchmarks/insert_tail_latency.cpp b/benchmarks/insert_tail_latency.cpp
new file mode 100644
index 0000000..5e32898
--- /dev/null
+++ b/benchmarks/insert_tail_latency.cpp
@@ -0,0 +1,80 @@
+/*
+ *
+ */
+
+#define ENABLE_TIMER
+
+#include <thread>
+
+#include "framework/DynamicExtension.h"
+#include "shard/ISAMTree.h"
+#include "query/rangecount.h"
+#include "framework/interface/Record.h"
+#include <unistd.h>
+#include <gsl/gsl_rng.h>
+
+#include "psu-util/timer.h"
+
+
+typedef de::Record<int64_t, int64_t> Rec;
+typedef de::ISAMTree<Rec> ISAM;
+typedef de::rc::Query<ISAM, Rec> Q;
+typedef de::DynamicExtension<Rec, ISAM, Q, de::LayoutPolicy::TEIRING, de::DeletePolicy::TAGGING, de::FIFOScheduler> Ext;
+
+std::atomic<size_t> total_latency = 0;
+
+void insert_thread(Ext *extension, size_t n, size_t k, size_t rate) {
+    int64_t delay = (1.0 / (double) rate) * 10e6; // delay in us
+    TIMER_INIT();
+    for (int64_t i=0; i<n; i+=k) {
+        TIMER_START();
+        for (int64_t j=0; j<k; j++) {
+            Rec r = {i+j, i+j};
+            while (!extension->insert(r)) {
+                _mm_pause();
+            }
+
+            //usleep(delay);
+            /*
+            for (size_t i=0; i<10000; i++) {
+                __asm__ __volatile__ ("":::"memory");
+            }
+            */
+        }
+        TIMER_STOP();
+
+        auto insert_lat =  TIMER_RESULT();
+
+        total_latency.fetch_add(insert_lat);
+        fprintf(stdout, "I\t%ld\t%ld\t%ld\n", i+k, insert_lat, k);
+    }
+}
+
+int main(int argc, char **argv) {
+
+    /* the closeout routine takes _forever_ ... so we'll just leak the memory */
+    auto extension = new Ext(100, 1000000, 3);
+    size_t n = 100000000;
+    size_t per_trial = 1000;
+    double selectivity = .001;
+    size_t rate = 1000000;
+
+    total_latency.store(0);
+
+    gsl_rng * rng = gsl_rng_alloc(gsl_rng_mt19937);
+
+    std::thread i_thrd1(insert_thread, extension, n/2, per_trial, rate);
+    std::thread i_thrd2(insert_thread, extension, n/2, per_trial, rate);
+
+    i_thrd1.join();
+    i_thrd2.join();
+
+    auto avg_latency = total_latency.load() / n;
+    auto throughput = (int64_t) ((double) n / (double) total_latency * 1e9);
+
+    fprintf(stdout, "AVG LAT: %ld\nThroughput: %ld\n", avg_latency, throughput);
+
+    gsl_rng_free(rng);
+    fflush(stderr);
+}
+
-- 
cgit v1.2.3


From d166465dcca3550cb8f3263e0f5b5189a69d531a Mon Sep 17 00:00:00 2001
From: Douglas Rumbaugh <dbr4@psu.edu>
Date: Wed, 31 Jan 2024 13:29:49 -0500
Subject: Temporary thread affinity for reconstruction

---
 benchmarks/insert_tail_latency.cpp | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

(limited to 'benchmarks/insert_tail_latency.cpp')

diff --git a/benchmarks/insert_tail_latency.cpp b/benchmarks/insert_tail_latency.cpp
index 5e32898..1640ce5 100644
--- a/benchmarks/insert_tail_latency.cpp
+++ b/benchmarks/insert_tail_latency.cpp
@@ -19,7 +19,7 @@
 typedef de::Record<int64_t, int64_t> Rec;
 typedef de::ISAMTree<Rec> ISAM;
 typedef de::rc::Query<ISAM, Rec> Q;
-typedef de::DynamicExtension<Rec, ISAM, Q, de::LayoutPolicy::TEIRING, de::DeletePolicy::TAGGING, de::FIFOScheduler> Ext;
+typedef de::DynamicExtension<Rec, ISAM, Q, de::LayoutPolicy::TEIRING, de::DeletePolicy::TAGGING, de::SerialScheduler> Ext;
 
 std::atomic<size_t> total_latency = 0;
 
@@ -53,8 +53,8 @@ void insert_thread(Ext *extension, size_t n, size_t k, size_t rate) {
 int main(int argc, char **argv) {
 
     /* the closeout routine takes _forever_ ... so we'll just leak the memory */
-    auto extension = new Ext(100, 1000000, 3);
-    size_t n = 100000000;
+    auto extension = new Ext(12000, 12001, 3);
+    size_t n = 10000000;
     size_t per_trial = 1000;
     double selectivity = .001;
     size_t rate = 1000000;
@@ -63,11 +63,12 @@ int main(int argc, char **argv) {
 
     gsl_rng * rng = gsl_rng_alloc(gsl_rng_mt19937);
 
-    std::thread i_thrd1(insert_thread, extension, n/2, per_trial, rate);
-    std::thread i_thrd2(insert_thread, extension, n/2, per_trial, rate);
+    std::thread i_thrd1(insert_thread, extension, n, per_trial, rate);
+    //std::thread i_thrd2(insert_thread, extension, n/2, per_trial, rate);
+
 
     i_thrd1.join();
-    i_thrd2.join();
+    //i_thrd2.join();
 
     auto avg_latency = total_latency.load() / n;
     auto throughput = (int64_t) ((double) n / (double) total_latency * 1e9);
-- 
cgit v1.2.3


From 2c5d549b3618b9ea72e6eece4cb4f3da5a6811a8 Mon Sep 17 00:00:00 2001
From: Douglas Rumbaugh <dbr4@psu.edu>
Date: Wed, 7 Feb 2024 13:42:34 -0500
Subject: Fully realized shard concept interface

---
 benchmarks/insert_tail_latency.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'benchmarks/insert_tail_latency.cpp')

diff --git a/benchmarks/insert_tail_latency.cpp b/benchmarks/insert_tail_latency.cpp
index 1640ce5..bdc4536 100644
--- a/benchmarks/insert_tail_latency.cpp
+++ b/benchmarks/insert_tail_latency.cpp
@@ -18,7 +18,7 @@
 
 typedef de::Record<int64_t, int64_t> Rec;
 typedef de::ISAMTree<Rec> ISAM;
-typedef de::rc::Query<ISAM, Rec> Q;
+typedef de::rc::Query<Rec, ISAM> Q;
 typedef de::DynamicExtension<Rec, ISAM, Q, de::LayoutPolicy::TEIRING, de::DeletePolicy::TAGGING, de::SerialScheduler> Ext;
 
 std::atomic<size_t> total_latency = 0;
-- 
cgit v1.2.3