From 7440266737210c7979178d1747cc3d68594f364f Mon Sep 17 00:00:00 2001 From: Douglas Rumbaugh Date: Wed, 9 Apr 2025 10:55:27 -0400 Subject: Disabled early abort for point lookups I'm having some trouble getting "bad" query performance, so I'm going to try using PLs w/o early abort as a worst-case scenario. This should get the best per-shard performance scaling --- benchmarks/include/file_util.h | 26 ++++++++++++++++++++++++ benchmarks/tail-latency/insert_query_threads.cpp | 17 ++++++++-------- include/query/pointlookup.h | 2 +- 3 files changed, 36 insertions(+), 9 deletions(-) diff --git a/benchmarks/include/file_util.h b/benchmarks/include/file_util.h index df8d999..a159dbe 100644 --- a/benchmarks/include/file_util.h +++ b/benchmarks/include/file_util.h @@ -10,6 +10,7 @@ #include #include #include +#include #include #include "psu-util/progress.h" @@ -73,6 +74,7 @@ static std::vector read_range_queries(std::string &fname, while (fscanf(qf, "%zu%zu%lf\n", &start, &stop, &sel) != EOF) { if (start < stop && std::abs(sel - selectivity) < 0.00001) { QP q; + q.lower_bound = start; q.upper_bound = stop; @@ -84,6 +86,30 @@ static std::vector read_range_queries(std::string &fname, return queries; } +template +static std::vector read_sosd_point_lookups(std::string &fname, size_t n) { + std::vector queries; + + FILE *qf = fopen(fname.c_str(), "r"); + + if (!qf) { + fprintf(stderr, "ERROR: Failed to open file %s\n", fname.c_str()); + exit(EXIT_FAILURE); + } + + size_t start, stop; + double sel; + while (fscanf(qf, "%zu%zu%lf\n", &start, &stop, &sel) != EOF && queries.size() < n) { + QP q; + + q.search_key= start; + queries.push_back(q); + } + fclose(qf); + + return queries; +} + template static std::vector read_binary_knn_queries(std::string fname, size_t k, size_t n) { diff --git a/benchmarks/tail-latency/insert_query_threads.cpp b/benchmarks/tail-latency/insert_query_threads.cpp index 0d92fde..d5939de 100644 --- a/benchmarks/tail-latency/insert_query_threads.cpp +++ b/benchmarks/tail-latency/insert_query_threads.cpp @@ -14,8 +14,8 @@ #include "framework/scheduling/FIFOScheduler.h" #include "framework/scheduling/SerialScheduler.h" #include "framework/util/Configuration.h" -#include "query/rangecount.h" -#include "shard/TrieSpline.h" +#include "query/pointlookup.h" +#include "shard/ISAMTree.h" #include "standard_benchmarks.h" #include "util/types.h" @@ -26,8 +26,8 @@ #include "psu-util/timer.h" typedef de::Record Rec; -typedef de::TrieSpline Shard; -typedef de::rc::Query Q; +typedef de::ISAMTree Shard; +typedef de::pl::Query Q; typedef de::DynamicExtension Ext; @@ -64,7 +64,7 @@ void query_thread(Ext *extension, std::vector *queries) { TIMER_STOP(); total_query_time.fetch_add(TIMER_RESULT()); - total_res.fetch_add(res); + total_res.fetch_add(res.size()); } } @@ -99,15 +99,16 @@ int main(int argc, char **argv) { std::string q_fname = std::string(argv[3]); auto data = read_sosd_file(d_fname, n); - auto queries = read_range_queries(q_fname, .0001); + //auto queries = read_range_queries(q_fname, .0001); + auto queries =read_sosd_point_lookups(q_fname, 100); std::vector sfs = {8}; //, 4, 8, 16, 32, 64, 128, 256, 512, 1024}; size_t buffer_size = 8000; - std::vector policies = {2}; + std::vector policies = {0, 1, 2}; std::vector thread_counts = {8}; std::vector modifiers = {0}; - std::vector scale_factors = {2, 3, 4, 5, 6, 7, 8}; + std::vector scale_factors = {2, 4, 8, 16, 32, 64, 128, 256}; size_t insert_threads = 1; size_t query_threads = 1; diff --git a/include/query/pointlookup.h b/include/query/pointlookup.h index 65cffa7..b42ec85 100644 --- a/include/query/pointlookup.h +++ b/include/query/pointlookup.h @@ -41,7 +41,7 @@ public: typedef std::vector> LocalResultType; typedef std::vector ResultType; - constexpr static bool EARLY_ABORT = true; + constexpr static bool EARLY_ABORT = false; constexpr static bool SKIP_DELETE_FILTER = true; static LocalQuery *local_preproc(S *shard, Parameters *parms) { -- cgit v1.2.3