add repeated experiments

qyliu-hkust · Aug 21, 2024 · 7ca4607 · 7ca4607
1 parent 2222f36
commit 7ca4607
Show file tree

Hide file tree

Showing 3 changed files with 125 additions and 86 deletions.
diff --git a/main.cpp b/main.cpp
@@ -17,8 +17,9 @@
 
 
 auto bench_search(const size_t& n, const size_t& nq) {
-    std::cout << "===========================================" << std::endl;
+    std::cout << "====== n=" << n << " nq=" << nq << " ======" << std::endl;
     auto data = benchmark::gen_random_keys<uint64_t>(n);
+    std::sort(data.begin(), data.end());
     auto queries = benchmark::gen_random_keys<uint64_t>(nq);
 
     volatile uint64_t res = 0;
@@ -31,18 +32,22 @@ auto bench_search(const size_t& n, const size_t& nq) {
     std::cout << "Search result " << res << std::endl;
     std::cout << "Query latency (linear) " << duration_linear / nq << std::endl;
 
+    std::vector<uint64_t> data_cpy1(data);
+    std::vector<uint64_t> queries_cpy1(queries);
     start = std::chrono::high_resolution_clock::now();
-    for (auto q : queries) {
-        res = *search::upper_bound_branchless(data.begin(), data.end(), q);
+    for (auto q : queries_cpy1) {
+        res = *search::upper_bound_branchless(data_cpy1.begin(), data_cpy1.end(), q);
     }
     end = std::chrono::high_resolution_clock::now();
     auto duration_branchless = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start).count();
     std::cout << "Search result " << res << std::endl;
     std::cout << "Query latency (branchless) " << duration_branchless / nq << std::endl;
 
+    std::vector<uint64_t> data_cpy2(data);
+    std::vector<uint64_t> queries_cpy2(queries);
     start = std::chrono::high_resolution_clock::now();
-    for (auto q : queries) {
-        res = *std::lower_bound(data.begin(), data.end(), q);
+    for (auto q : queries_cpy2) {
+        res = *std::lower_bound(data_cpy2.begin(), data_cpy2.end(), q);
     }
     end = std::chrono::high_resolution_clock::now();
     auto duration_branchy = std::chrono::duration_cast<std::chrono::nanoseconds>(end - start).count();
@@ -53,6 +58,26 @@ auto bench_search(const size_t& n, const size_t& nq) {
     return timer {duration_branchless/nq, duration_branchy/nq, duration_linear/nq};
 }
 
+void bench_search_repeat(const size_t& repeat, const size_t& nq, const std::string& ouput_fname) {
+    const size_t n_max = 21;
+    size_t n_list[n_max];
+    for (auto i=1; i<=21; ++i) {
+        n_list[i-1] = pow(2, i);
+    }
+
+    std::ofstream ofs(ouput_fname);
+    ofs << "n,round,brl,br,linear" << std::endl;
+
+    for (auto n : n_list) {
+        for (auto i=0; i<repeat; ++i) {
+            auto res = bench_search(n, nq);
+            ofs << n << "," << i << "," << res.brl << "," << res.br << "," << res.linear << std::endl;
+        }
+    }
+
+    ofs.close();
+}
+
 struct stats {
     size_t eps_l;
     size_t eps_i;
@@ -66,6 +91,7 @@ struct stats {
     size_t latency_branchless_l;
 };
 
+
 template<size_t Epsilon, size_t EpsilonRecursive>
 auto bench_pgm(const std::vector<uint64_t>& data, const std::vector<uint64_t>& queries) {
     std::cout << "===========================================" << std::endl;
@@ -136,91 +162,100 @@ auto bench_pgm(const std::vector<uint64_t>& data, const std::vector<uint64_t>& q
 
 
 int main(int argc, const char * argv[]) {
+//    bench_search_repeat(20, 500, "/Users/liuqiyu/Desktop/bench_search_result_new.csv");
+//    exit(0);
+
     const std::string fname = "/Users/liuqiyu/Desktop/SOSD_data/fb_200M_uint64";
-    const size_t nq = 5000;
+    const size_t nq = 100;
+    const size_t repeat = 10;
 
     std::cout << "Load data from " << fname << std::endl;
     auto data = benchmark::load_data<uint64_t>(fname);
     std::sort(data.begin(), data.end());
 
-    std::cout << "Generate " << nq << " random search keys." << std::endl;
-    auto queries = benchmark::gen_random_keys<uint64_t>(nq);
+    std::vector<std::pair<size_t, stats>> bench_results;
 
-    std::vector<stats> bench_results;
-    bench_results.emplace_back(bench_pgm<4, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 4>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 4>(data, queries));
-
-    bench_results.emplace_back(bench_pgm<4, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 8>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 8>(data, queries));
-
-    bench_results.emplace_back(bench_pgm<4, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 16>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 16>(data, queries));
-
-    bench_results.emplace_back(bench_pgm<4, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 32>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 32>(data, queries));
-
-    bench_results.emplace_back(bench_pgm<4, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 64>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 64>(data, queries));
-
-    bench_results.emplace_back(bench_pgm<4, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<8, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<16, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<32, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<64, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<128, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<256, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<512, 128>(data, queries));
-    bench_results.emplace_back(bench_pgm<1024, 128>(data, queries));
+    for (auto i=0; i<repeat; ++i) {
+        std::cout << "Round " << i << std::endl;
+        std::cout << "Generate " << nq << " random search keys." << std::endl;
+        auto queries = benchmark::gen_random_keys<uint64_t>(nq);
+
+        bench_results.emplace_back(i, bench_pgm<4, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 4>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 4>(data, queries));
+
+        bench_results.emplace_back(i, bench_pgm<4, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 8>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 8>(data, queries));
+
+        bench_results.emplace_back(i, bench_pgm<4, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 16>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 16>(data, queries));
+
+        bench_results.emplace_back(i, bench_pgm<4, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 32>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 32>(data, queries));
+
+        bench_results.emplace_back(i, bench_pgm<4, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 64>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 64>(data, queries));
+
+        bench_results.emplace_back(i, bench_pgm<4, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<8, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<16, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<32, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<64, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<128, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<256, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<512, 128>(data, queries));
+        bench_results.emplace_back(i, bench_pgm<1024, 128>(data, queries));
+    }
 
     // start from 7 cold cache config
-    std::ofstream ofs("/Users/liuqiyu/Desktop/bench_pgm_result_new_2.csv");
-    ofs << "eps_l,eps_i,levels,lls,ils,latency_branchy_i,latency_branchy_l,latency_branchless_i,latency_branchless_l" << std::endl;
+    std::ofstream ofs("/Users/liuqiyu/Desktop/bench_pgm_result_new_repeat_10.csv");
+    ofs << "round,eps_l,eps_i,levels,lls,ils,latency_branchy_i,latency_branchy_l,latency_branchless_i,latency_branchless_l" << std::endl;
 
     for (auto br : bench_results) {
-        ofs << br.eps_l << ","
-            << br.eps_i << ","
-            << br.levels << ","
-            << br.lls << ","
-            << br.ils << ","
-            << br.latency_branchy_i << "," 
-            << br.latency_branchy_l << ","
-            << br.latency_branchless_i << ","
-            << br.latency_branchless_l << std::endl;
+        ofs << br.first << ","
+            << br.second.eps_l << ","
+            << br.second.eps_i << ","
+            << br.second.levels << ","
+            << br.second.lls << ","
+            << br.second.ils << ","
+            << br.second.latency_branchy_i << ","
+            << br.second.latency_branchy_l << ","
+            << br.second.latency_branchless_i << ","
+            << br.second.latency_branchless_l << std::endl;
     }
 
 

diff --git a/pgm_index.h b/pgm_index.h
@@ -208,11 +208,15 @@ class PGMIndex {
 
     template<typename RandomIt>
     auto search_data(RandomIt start, const K &key) {
-        auto pos = search(key);
+        auto k = std::max(first_key, key);
+        auto it = segment_for_key(k);
+        auto pos = std::min<size_t>((*it)(k), std::next(it)->intercept);
+        auto lo = PGM_SUB_EPS(pos, Epsilon);
+        auto hi = PGM_ADD_EPS(pos, Epsilon, n);
         if constexpr (BranchLessSearch) {
-            return search::lower_bound_branchless(start+pos.lo, start+pos.hi, key);
+            return search::lower_bound_branchless(start+lo, start+hi, key);
         } else {
-            return std::lower_bound(start+pos.lo, start+pos.hi, key);
+            return std::lower_bound(start+lo, start+hi, key);
         }
     }
 

diff --git a/search_algo.h b/search_algo.h
@@ -15,11 +15,11 @@ template<typename RandomIt, typename K>
 RandomIt lower_bound_linear(RandomIt start, RandomIt end, const K& key) {
     auto it = start;
     while (it != end) {
-        if (*it ==key)
-            break;
-        it++;
+        if (*it >= key)
+            return it;
+        ++it;
     }
-    return it;
+    return end;
 }