Benchmarking programs

14 files changed, 1343 insertions, 0 deletions

diff --git a/benchmarks/old-bench/alex_rq_bench.cpp b/benchmarks/old-bench/alex_rq_bench.cpp
new file mode 100644
index 0000000..f75afa6
--- /dev/null
+++ b/benchmarks/old-bench/alex_rq_bench.cpp
@@ -0,0 +1,205 @@
+#include "alex.h"
+#include "include/standalone_utility.h"
+
+typedef uint64_t key_type;
+typedef uint64_t value_type;
+
+typedef alex::Alex<key_type, value_type> Alex;
+
+struct record {
+    key_type key;
+    value_type value;
+};
+
+struct query {
+    key_type lower_bound;
+    key_type upper_bound;
+};
+
+template <typename R>
+static bool build_insert_vec(std::fstream &file, std::vector<R> &vec, size_t n, 
+                             double delete_prop, std::vector<R> &to_delete, bool binary=false) {
+    vec.clear();
+    for (size_t i=0; i<n; i++) {
+        R rec;
+        if (!next_record(file, rec, binary)) {
+            if (i == 0) {
+                return false;
+            }
+
+            break;
+        }
+
+        vec.emplace_back(rec);
+
+        if (gsl_rng_uniform(g_rng) < delete_prop + (delete_prop * .1)) {
+            to_delete.emplace_back(rec);
+        }
+    }
+
+    return true;
+}
+
+
+static Alex *warmup(std::fstream &file, size_t count, 
+                   double delete_prop, std::vector<record> to_delete, bool progress=true, bool binary=false) {
+    size_t batch = std::min(.1 * count, 25000.0);
+
+    std::pair<key_type, value_type> *insert_vec = new std::pair<key_type, value_type>[count];
+    Alex *alex = new Alex();
+
+    size_t cnt = 0;
+    record rec;
+    while (cnt < count && next_record(file, rec)) {
+        insert_vec[cnt] = {rec.key, rec.value};
+        cnt++;
+    }
+
+    std::sort(insert_vec, insert_vec + count);
+
+    alex->bulk_load(insert_vec, count); 
+    delete[] insert_vec;
+
+    return alex;
+}
+
+
+static void alex_rq_insert(Alex &alex, std::fstream &file, size_t insert_cnt, double delete_prop, std::vector<record> &to_delete, bool binary=false) {
+    size_t delete_cnt = insert_cnt * delete_prop;
+
+    size_t applied_deletes = 0;
+    size_t applied_inserts = 0;
+
+    size_t BATCH=1000;
+
+    std::vector<record> insert_vec;
+    std::vector<record> delete_vec;
+    insert_vec.reserve(BATCH);
+    delete_vec.reserve(BATCH*delete_prop);
+
+    size_t delete_idx = 0;
+
+    bool continue_benchmark = true;
+
+    size_t total_time = 0;
+
+    while (applied_inserts < insert_cnt && continue_benchmark) { 
+        continue_benchmark = build_insert_vec(file, insert_vec, BATCH, delete_prop, to_delete, binary);
+        progress_update((double) applied_inserts / (double) insert_cnt, "inserting:");
+        if (applied_deletes < delete_cnt) {
+            build_delete_vec(to_delete, delete_vec, BATCH*delete_prop);
+            delete_idx = 0;
+        }
+
+        if (insert_vec.size() == 0) {
+            break;
+        }
+
+        auto insert_start = std::chrono::high_resolution_clock::now();
+        for (size_t i=0; i<insert_vec.size(); i++) {
+            // process a delete if necessary
+            if (applied_deletes < delete_cnt && delete_idx < delete_vec.size() && gsl_rng_uniform(g_rng) < delete_prop) {
+                alex.erase_one(delete_vec[delete_idx++].key);
+                applied_deletes++;
+            }
+
+            // insert the record;
+            alex.insert(insert_vec[i].key, insert_vec[i].value);
+            applied_inserts++;
+        }
+        auto insert_stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(insert_stop - insert_start).count();
+    } 
+
+    progress_update(1.0, "inserting:");
+
+    size_t throughput = (((double) (applied_inserts + applied_deletes) / (double) total_time) * 1e9);
+
+    fprintf(stdout, "%ld\t", throughput);
+}
+
+
+
+static void alex_rq_bench(Alex &alex, std::vector<query> queries, size_t trial_cnt=1) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    std::vector<record> result_set;
+
+    for (int i=0; i<trial_cnt; i++) {
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            auto ptr = alex.find(queries[j].lower_bound);
+            while (ptr != alex.end() && ptr.key() <= queries[j].upper_bound) {
+                result_set.push_back({ptr.key(), ptr.payload()});
+                ptr++;
+            }
+            result_set.clear();
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: alex_rq_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+    bool use_osm = false;
+
+    double insert_batch = 0.8; 
+
+    init_bench_env(record_count, true, use_osm);
+    auto queries = read_range_queries<query>(qfilename, .0001);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<record> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    auto alex = warmup(datafile, warmup_cnt, delete_prop, to_delete, true, true);
+
+    fprintf(stderr, "Size: %ld\n", alex->size());
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    alex_rq_insert(*alex, datafile, insert_cnt, delete_prop, to_delete, true);
+    size_t memory_usage = alex->model_size() + alex->data_size();
+
+    fprintf(stderr, "Size: %ld\n", alex->size());
+    fprintf(stdout, "%ld\t", memory_usage);
+
+    alex_rq_bench(*alex, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    delete alex;
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/alias_wss_bench.cpp b/benchmarks/old-bench/alias_wss_bench.cpp
new file mode 100644
index 0000000..a3a43f2
--- /dev/null
+++ b/benchmarks/old-bench/alias_wss_bench.cpp
@@ -0,0 +1,57 @@
+/*
+ * benchmarks/alias_wss_bench.cpp
+ *
+ * Copyright (C) 2023 Douglas Rumbaugh <drumbaugh@psu.edu> 
+ *
+ * All rights reserved. Published under the Modified BSD License.
+ *
+ */
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 4) {
+        fprintf(stderr, "Usage: sampling_tput <filename> <record_count> <delete_proportion> [osm_data]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double delete_prop = atof(argv[3]);
+    double max_delete_prop = (delete_prop > 0) ? delete_prop : 1;
+    bool use_osm = (argc == 5) ? atoi(argv[4]) : 0;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+
+    auto de_wss = ExtendedWSS(buffer_cap, scale_factor, max_delete_prop);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in);
+
+    std::vector<WRec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedWSS, WRec>(datafile, de_wss, warmup_cnt, delete_prop, to_delete);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    std::vector<de::wss_query_parms<WRec>> queries(1);
+    queries[0].rng = g_rng;
+    queries[0].sample_size = 1000;
+
+    insert_tput_bench<ExtendedWSS, WRec>(de_wss, datafile, insert_cnt, delete_prop, to_delete);
+    query_latency_bench<ExtendedWSS, WRec, de::wss_query_parms<WRec>>(de_wss, queries, 1000);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/btree_irs_bench.cpp b/benchmarks/old-bench/btree_irs_bench.cpp
new file mode 100644
index 0000000..862fc6b
--- /dev/null
+++ b/benchmarks/old-bench/btree_irs_bench.cpp
@@ -0,0 +1,91 @@
+#include "include/bench.h"
+#include "ds/BTree.h"
+
+static void btree_sample_bench(TreeMap &tree, std::vector<de::irs_query_parms<btree_record>> queries, size_t trial_cnt=10) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    std::vector<key_type> sample_set;
+    sample_set.reserve(queries[0].sample_size);
+
+    for (int i=0; i<trial_cnt; i++) {
+        progress_update((double) (i * batch_size) / (double) trial_cnt, progbuf);
+
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            tree.range_sample(queries[j].lower_bound, queries[j].upper_bound, queries[j].sample_size, sample_set, g_rng);
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    progress_update(1.0, progbuf);
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: btree_irs_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+    bool use_osm = false;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+    auto queries = read_range_queries<de::irs_query_parms<btree_record>>(qfilename, .001);
+
+    for (auto &q: queries) {
+        q.rng = g_rng;
+        q.sample_size = 1000;
+    }
+
+    auto btree = TreeMap();
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<btree_record> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<TreeMap, btree_record>(datafile, btree, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<TreeMap, btree_record>(btree, datafile, insert_cnt, delete_prop, to_delete, true);
+    size_t memory_usage = btree.get_stats().inner_nodes * tlx::btree_default_traits<key_type, btree_record>::inner_slots * (sizeof(key_type) + sizeof(void*));
+    memory_usage += btree.get_stats().leaves * tlx::btree_default_traits<key_type, btree_record>::leaf_slots * sizeof(btree_record);
+    fprintf(stdout, "%ld\t", memory_usage);
+
+    btree_sample_bench(btree, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/btree_rq_bench.cpp b/benchmarks/old-bench/btree_rq_bench.cpp
new file mode 100644
index 0000000..d92b45d
--- /dev/null
+++ b/benchmarks/old-bench/btree_rq_bench.cpp
@@ -0,0 +1,90 @@
+#include "include/bench.h"
+#include "ds/BTree.h"
+
+static void btree_rq_bench(TreeMap &tree, std::vector<de::ISAMRangeQueryParms<btree_record>> queries, size_t trial_cnt=1) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    std::vector<btree_record> result_set;
+
+    for (int i=0; i<trial_cnt; i++) {
+        progress_update((double) (i * batch_size) / (double) trial_cnt, progbuf);
+
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            auto ptr = tree.find(queries[j].lower_bound);
+            while (ptr != tree.end() && ptr->key <= queries[j].upper_bound) {
+                result_set.emplace_back(*ptr);
+                ptr++;
+            }
+            result_set.clear();
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    progress_update(1.0, progbuf);
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: btree_rq_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+    bool use_osm = false;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+    auto queries = read_range_queries<de::ISAMRangeQueryParms<btree_record>>(qfilename, .0001);
+
+    auto btree = TreeMap();
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<btree_record> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<TreeMap, btree_record>(datafile, btree, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<TreeMap, btree_record>(btree, datafile, insert_cnt, delete_prop, to_delete, true);
+    size_t memory_usage = btree.get_stats().inner_nodes * tlx::btree_default_traits<key_type, btree_record>::inner_slots * (sizeof(key_type) + sizeof(void*));
+    memory_usage += btree.get_stats().leaves * tlx::btree_default_traits<key_type, btree_record>::leaf_slots * sizeof(btree_record);
+    fprintf(stdout, "%ld\t", memory_usage);
+
+    btree_rq_bench(btree, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/isam_irs_bench.cpp b/benchmarks/old-bench/isam_irs_bench.cpp
new file mode 100644
index 0000000..96525f0
--- /dev/null
+++ b/benchmarks/old-bench/isam_irs_bench.cpp
@@ -0,0 +1,64 @@
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: isam_irs_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+    bool use_osm = false;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+    auto queries = read_range_queries<de::irs_query_parms<Rec>>(qfilename, .001);
+
+    for (auto &q: queries) {
+        q.rng = g_rng;
+        q.sample_size = 1000;
+    }
+
+    auto de_irs = ExtendedISAM_IRS(buffer_cap, scale_factor, max_delete_prop);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Rec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedISAM_IRS, Rec>(datafile, de_irs, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedISAM_IRS, Rec>(de_irs, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de_irs.get_memory_usage());
+    query_latency_bench<ExtendedISAM_IRS, Rec, de::irs_query_parms<Rec>>(de_irs, queries);
+    fprintf(stdout, "\n");
+
+    auto ts = de_irs.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::MemISAM<Rec>, Rec, de::irs_query_parms<Rec>, de::IRSQuery<Rec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/isam_rq_bench.cpp b/benchmarks/old-bench/isam_rq_bench.cpp
new file mode 100644
index 0000000..bb5626e
--- /dev/null
+++ b/benchmarks/old-bench/isam_rq_bench.cpp
@@ -0,0 +1,59 @@
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: isam_rq_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+    bool use_osm = false;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+    auto queries = read_range_queries<de::ISAMRangeQueryParms<Rec>>(qfilename, .0001);
+
+    auto de_isam_rq = ExtendedISAM_RQ(buffer_cap, scale_factor, max_delete_prop);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Rec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedISAM_RQ, Rec>(datafile, de_isam_rq, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedISAM_RQ, Rec>(de_isam_rq, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de_isam_rq.get_memory_usage());
+    query_latency_bench<ExtendedISAM_RQ, Rec, de::ISAMRangeQueryParms<Rec>>(de_isam_rq, queries);
+    fprintf(stdout, "\n");
+
+    auto ts = de_isam_rq.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::MemISAM<Rec>, Rec, de::ISAMRangeQueryParms<Rec>, de::ISAMRangeQuery<Rec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/mtree_knn_bench.cpp b/benchmarks/old-bench/mtree_knn_bench.cpp
new file mode 100644
index 0000000..9d4cc57
--- /dev/null
+++ b/benchmarks/old-bench/mtree_knn_bench.cpp
@@ -0,0 +1,83 @@
+#include "include/bench.h"
+#include "mtree.h"
+
+static void mtree_knn_bench(MTree &tree, std::vector<de::KNNQueryParms<Word2VecRec>> queries, size_t trial_cnt=1) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    std::vector<Word2VecRec> result_set;
+
+    for (int i=0; i<trial_cnt; i++) {
+        progress_update((double) (i * batch_size) / (double) trial_cnt, progbuf);
+
+        std::vector<Word2VecRec> results;
+
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            results.clear();
+            auto query_output = tree.get_nearest_by_limit(queries[j].point, queries[j].k);
+            auto itr = query_output.begin();
+            while (itr != query_output.end()) {
+                results.emplace_back(itr->data);
+                itr++;
+            }
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    progress_update(1.0, progbuf);
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: mtree_knn_bench <filename> <record_count> <delete_proportion> <query_file> [k]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+    size_t k = (argc == 6) ? atol(argv[5]) : 10;
+
+    init_bench_env(record_count, true);
+    auto queries = read_knn_queries<de::KNNQueryParms<Word2VecRec>>(qfilename, k);
+
+    auto mtree = MTree();
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Word2VecRec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = 0.1 * record_count;
+    warmup<MTree, Word2VecRec>(datafile, mtree, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<MTree, Word2VecRec>(mtree, datafile, insert_cnt, delete_prop, to_delete, true);
+   // fprintf(stdout, "%ld\t", mtree.get_memory_usage());
+
+    mtree_knn_bench(mtree, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/pgm_pl_bench.cpp b/benchmarks/old-bench/pgm_pl_bench.cpp
new file mode 100644
index 0000000..f798861
--- /dev/null
+++ b/benchmarks/old-bench/pgm_pl_bench.cpp
@@ -0,0 +1,67 @@
+/*
+ * benchmarks/triespline_rq_bench.cpp
+ *
+ * Copyright (C) 2023 Douglas Rumbaugh <drumbaugh@psu.edu> 
+ *
+ * All rights reserved. Published under the Modified BSD License.
+ *
+ */
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: pgm_pl_bench <filename> <record_count> <delete_proportion> <query_file> [osm_data]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    size_t buffer_cap = 1000;
+    size_t scale_factor = 6;
+    double delete_prop = atof(argv[3]);
+    double max_delete_prop = (delete_prop > 0) ? delete_prop : 1;
+    std::string query_file = std::string(argv[4]);
+    bool use_osm = (argc == 6) ? atoi(argv[5]) : 0;
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true, use_osm);
+
+    auto de = ExtendedPGM_PL(buffer_cap, scale_factor, max_delete_prop);
+    auto queries = read_lookup_queries<de::PGMPointLookupParms<Rec>>(query_file, .0001);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Rec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedPGM_PL, Rec>(datafile, de, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedPGM_PL, Rec>(de, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de.get_memory_usage());
+    query_latency_bench<ExtendedPGM_PL, Rec, de::PGMPointLookupParms<Rec>>(de, queries, 1);
+
+    fprintf(stdout, "\n");
+
+    auto ts = de.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::PGM<Rec>, Rec, de::PGMPointLookupParms<Rec>, de::PGMPointLookup<Rec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/pgm_rq_bench.cpp b/benchmarks/old-bench/pgm_rq_bench.cpp
new file mode 100644
index 0000000..e25d29f
--- /dev/null
+++ b/benchmarks/old-bench/pgm_rq_bench.cpp
@@ -0,0 +1,67 @@
+/*
+ * benchmarks/triespline_rq_bench.cpp
+ *
+ * Copyright (C) 2023 Douglas Rumbaugh <drumbaugh@psu.edu> 
+ *
+ * All rights reserved. Published under the Modified BSD License.
+ *
+ */
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: pgm_rq_bench <filename> <record_count> <delete_proportion> <query_file> [osm_data]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 8;
+    double delete_prop = atof(argv[3]);
+    double max_delete_prop = (delete_prop > 0) ? delete_prop : 1;
+    std::string query_file = std::string(argv[4]);
+    bool use_osm = (argc == 6) ? atoi(argv[5]) : 0;
+
+    double insert_batch = 0.5; 
+
+    init_bench_env(record_count, true, use_osm);
+
+    auto de = ExtendedPGMRQ(buffer_cap, scale_factor, max_delete_prop);
+    auto queries = read_range_queries<de::pgm_range_query_parms<Rec>>(query_file, .0001);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Rec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedPGMRQ, Rec>(datafile, de, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedPGMRQ, Rec>(de, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de.get_memory_usage());
+    query_latency_bench<ExtendedPGMRQ, Rec, de::pgm_range_query_parms<Rec>>(de, queries, 1);
+
+    fprintf(stdout, "\n");
+
+    auto ts = de.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::PGM<Rec>, Rec, de::pgm_range_query_parms<Rec>, de::PGMRangeQuery<Rec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/test.cpp b/benchmarks/old-bench/test.cpp
new file mode 100644
index 0000000..75bffe3
--- /dev/null
+++ b/benchmarks/old-bench/test.cpp
@@ -0,0 +1,7 @@
+#include "alex.h"
+
+
+int main(int argc, char **argv) {
+    alex::Alex<int, int> test;
+
+}
diff --git a/benchmarks/old-bench/triespline_rq_bench.cpp b/benchmarks/old-bench/triespline_rq_bench.cpp
new file mode 100644
index 0000000..967c3b0
--- /dev/null
+++ b/benchmarks/old-bench/triespline_rq_bench.cpp
@@ -0,0 +1,66 @@
+/*
+ * benchmarks/triespline_rq_bench.cpp
+ *
+ * Copyright (C) 2023 Douglas Rumbaugh <drumbaugh@psu.edu> 
+ *
+ * All rights reserved. Published under the Modified BSD License.
+ *
+ */
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: triespline_rq_bench <filename> <record_count> <delete_proportion> <query_file> [osm_data]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 8;
+    double delete_prop = atof(argv[3]);
+    double max_delete_prop = (delete_prop > 0) ? delete_prop : 1;
+    std::string query_file = std::string(argv[4]);
+    bool use_osm = (argc == 6) ? atoi(argv[5]) : 0;
+
+    double insert_batch = 0.5; 
+
+    init_bench_env(record_count, true, use_osm);
+
+    auto de = ExtendedTSRQ(buffer_cap, scale_factor, max_delete_prop);
+    auto queries = read_range_queries<de::ts_range_query_parms<Rec>>(query_file, .0001);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Rec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup<ExtendedTSRQ, Rec>(datafile, de, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedTSRQ, Rec>(de, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de.get_memory_usage());
+    query_latency_bench<ExtendedTSRQ, Rec, de::ts_range_query_parms<Rec>>(de, queries, 1);
+    fprintf(stdout, "\n");
+
+    auto ts = de.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::TrieSpline<Rec>, Rec, de::ts_range_query_parms<Rec>, de::TrieSplineRangeQuery<Rec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/upgm_pl_bench.cpp b/benchmarks/old-bench/upgm_pl_bench.cpp
new file mode 100644
index 0000000..e0445b2
--- /dev/null
+++ b/benchmarks/old-bench/upgm_pl_bench.cpp
@@ -0,0 +1,212 @@
+#include "pgm/pgm_index_dynamic.hpp"
+#include "include/standalone_utility.h"
+
+typedef uint64_t key_type;
+typedef uint64_t value_type;
+
+typedef pgm::DynamicPGMIndex<key_type, value_type, pgm::PGMIndex<key_type, 64>> PGM;
+
+struct record {
+    key_type key;
+    value_type value;
+};
+
+struct query {
+    key_type lower_bound;
+    key_type upper_bound;
+};
+
+template <typename R>
+static bool build_insert_vec(std::fstream &file, std::vector<R> &vec, size_t n, 
+                             double delete_prop, std::vector<R> &to_delete, bool binary=false) {
+    vec.clear();
+    for (size_t i=0; i<n; i++) {
+        R rec;
+        if (!next_record(file, rec, binary)) {
+            if (i == 0) {
+                return false;
+            }
+
+            break;
+        }
+
+        vec.emplace_back(rec);
+
+        if (gsl_rng_uniform(g_rng) < delete_prop + (delete_prop * .1)) {
+            to_delete.emplace_back(rec);
+        }
+    }
+
+    return true;
+}
+
+
+static bool warmup(std::fstream &file, PGM &pgm, size_t count, 
+                   double delete_prop, std::vector<record> to_delete, bool progress=true, bool binary=false) {
+    size_t batch = std::min(.1 * count, 25000.0);
+
+    std::vector<record> insert_vec;
+    std::vector<record> delete_vec;
+    insert_vec.reserve(batch);
+    delete_vec.reserve(batch*delete_prop);
+
+    size_t inserted = 0;
+    size_t delete_idx = 0;
+
+    double last_percent = 0;
+    while (inserted < count) {
+        // Build vector of records to insert and potentially delete
+        auto continue_warmup = build_insert_vec<record>(file, insert_vec, batch, delete_prop, to_delete, binary);
+        if (inserted > batch) {
+            build_delete_vec(to_delete, delete_vec, batch*delete_prop);
+            delete_idx = 0;
+        }
+
+        for (size_t i=0; i<insert_vec.size(); i++) {
+            // process a delete if necessary
+            if (delete_idx < delete_vec.size() && gsl_rng_uniform(g_rng) < delete_prop) {
+                pgm.erase(delete_vec[delete_idx++].key);
+            }
+
+            pgm.insert_or_assign(insert_vec[i].key, insert_vec[i].value);
+            inserted++;
+            progress_update((double) inserted / (double) count, "warming up:");
+        }
+    }
+
+    return true;
+}
+
+
+static void pgm_rq_insert(PGM &pgm, std::fstream &file, size_t insert_cnt, double delete_prop, std::vector<record> &to_delete, bool binary=false) {
+    size_t delete_cnt = insert_cnt * delete_prop;
+
+    size_t applied_deletes = 0;
+    size_t applied_inserts = 0;
+
+    size_t BATCH=1000;
+
+    std::vector<record> insert_vec;
+    std::vector<record> delete_vec;
+    insert_vec.reserve(BATCH);
+    delete_vec.reserve(BATCH*delete_prop);
+
+    size_t delete_idx = 0;
+
+    bool continue_benchmark = true;
+
+    size_t total_time = 0;
+
+    while (applied_inserts < insert_cnt && continue_benchmark) { 
+        continue_benchmark = build_insert_vec(file, insert_vec, BATCH, delete_prop, to_delete, binary);
+        progress_update((double) applied_inserts / (double) insert_cnt, "inserting:");
+        if (applied_deletes < delete_cnt) {
+            build_delete_vec(to_delete, delete_vec, BATCH*delete_prop);
+            delete_idx = 0;
+        }
+
+        if (insert_vec.size() == 0) {
+            break;
+        }
+
+        auto insert_start = std::chrono::high_resolution_clock::now();
+        for (size_t i=0; i<insert_vec.size(); i++) {
+            // process a delete if necessary
+            if (applied_deletes < delete_cnt && delete_idx < delete_vec.size() && gsl_rng_uniform(g_rng) < delete_prop) {
+                pgm.erase(delete_vec[delete_idx++].key);
+                applied_deletes++;
+            }
+
+            // insert the record;
+            pgm.insert_or_assign(insert_vec[i].key, insert_vec[i].value);
+            applied_inserts++;
+        }
+        auto insert_stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(insert_stop - insert_start).count();
+    } 
+
+    progress_update(1.0, "inserting:");
+
+    size_t throughput = (((double) (applied_inserts + applied_deletes) / (double) total_time) * 1e9);
+
+    fprintf(stdout, "%ld\t", throughput);
+}
+
+
+
+static void pgm_pl_bench(PGM &pgm, std::vector<query> queries, size_t trial_cnt=1) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    std::vector<record> result_set;
+
+    for (int i=0; i<trial_cnt; i++) {
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            auto ptr = pgm.find(queries[j].lower_bound);
+            if (ptr != pgm.end() && ptr->first == queries[j].lower_bound) {
+                result_set.push_back({ptr->first, ptr->second});
+            }
+            result_set.clear();
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: upgm_pl_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    double insert_batch = 0.1; 
+
+    init_bench_env(record_count, true);
+    auto queries = read_range_queries<query>(qfilename, .0001);
+
+    std::vector<std::pair<key_type, value_type>> data;
+    PGM pgm(data.begin(), data.end());
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<record> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup(datafile, pgm, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    pgm_rq_insert(pgm, datafile, insert_cnt, delete_prop, to_delete, true);
+    size_t memory_usage = pgm.size_in_bytes();
+    fprintf(stdout, "%ld\t", memory_usage);
+
+    pgm_pl_bench(pgm, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/upgm_rq_bench.cpp b/benchmarks/old-bench/upgm_rq_bench.cpp
new file mode 100644
index 0000000..940a9e6
--- /dev/null
+++ b/benchmarks/old-bench/upgm_rq_bench.cpp
@@ -0,0 +1,217 @@
+#include "pgm/pgm_index_dynamic.hpp"
+#include "include/standalone_utility.h"
+
+typedef uint64_t key_type;
+typedef uint64_t value_type;
+
+typedef pgm::DynamicPGMIndex<key_type, value_type, pgm::PGMIndex<key_type, 64>> PGM;
+
+struct record {
+    key_type key;
+    value_type value;
+};
+
+struct query {
+    key_type lower_bound;
+    key_type upper_bound;
+};
+
+template <typename R>
+static bool build_insert_vec(std::fstream &file, std::vector<R> &vec, size_t n, 
+                             double delete_prop, std::vector<R> &to_delete, bool binary=false) {
+    vec.clear();
+    for (size_t i=0; i<n; i++) {
+        R rec;
+        if (!next_record(file, rec, binary)) {
+            if (i == 0) {
+                return false;
+            }
+
+            break;
+        }
+
+        vec.emplace_back(rec);
+
+        if (gsl_rng_uniform(g_rng) < delete_prop + (delete_prop * .1)) {
+            to_delete.emplace_back(rec);
+        }
+    }
+
+    return true;
+}
+
+
+static bool warmup(std::fstream &file, PGM &pgm, size_t count, 
+                   double delete_prop, std::vector<record> to_delete, bool progress=true, bool binary=false) {
+    size_t batch = std::min(.1 * count, 25000.0);
+
+    std::vector<record> insert_vec;
+    std::vector<record> delete_vec;
+    insert_vec.reserve(batch);
+    delete_vec.reserve(batch*delete_prop);
+
+    size_t inserted = 0;
+    size_t delete_idx = 0;
+
+    double last_percent = 0;
+    while (inserted < count) {
+        // Build vector of records to insert and potentially delete
+        auto continue_warmup = build_insert_vec<record>(file, insert_vec, batch, delete_prop, to_delete, binary);
+        if (inserted > batch) {
+            build_delete_vec(to_delete, delete_vec, batch*delete_prop);
+            delete_idx = 0;
+        }
+
+        for (size_t i=0; i<insert_vec.size(); i++) {
+            // process a delete if necessary
+            if (delete_idx < delete_vec.size() && gsl_rng_uniform(g_rng) < delete_prop) {
+                pgm.erase(delete_vec[delete_idx++].key);
+            }
+
+            pgm.insert_or_assign(insert_vec[i].key, insert_vec[i].value);
+            inserted++;
+            progress_update((double) inserted / (double) count, "warming up:");
+        }
+    }
+
+    return true;
+}
+
+
+static void pgm_rq_insert(PGM &pgm, std::fstream &file, size_t insert_cnt, double delete_prop, std::vector<record> &to_delete, bool binary=false) {
+    size_t delete_cnt = insert_cnt * delete_prop;
+
+    size_t applied_deletes = 0;
+    size_t applied_inserts = 0;
+
+    size_t BATCH=1000;
+
+    std::vector<record> insert_vec;
+    std::vector<record> delete_vec;
+    insert_vec.reserve(BATCH);
+    delete_vec.reserve(BATCH*delete_prop);
+
+    size_t delete_idx = 0;
+
+    bool continue_benchmark = true;
+
+    size_t total_time = 0;
+
+    while (applied_inserts < insert_cnt && continue_benchmark) { 
+        continue_benchmark = build_insert_vec(file, insert_vec, BATCH, delete_prop, to_delete, binary);
+        progress_update((double) applied_inserts / (double) insert_cnt, "inserting:");
+        if (applied_deletes < delete_cnt) {
+            build_delete_vec(to_delete, delete_vec, BATCH*delete_prop);
+            delete_idx = 0;
+        }
+
+        if (insert_vec.size() == 0) {
+            break;
+        }
+
+        auto insert_start = std::chrono::high_resolution_clock::now();
+        for (size_t i=0; i<insert_vec.size(); i++) {
+            // process a delete if necessary
+            if (applied_deletes < delete_cnt && delete_idx < delete_vec.size() && gsl_rng_uniform(g_rng) < delete_prop) {
+                pgm.erase(delete_vec[delete_idx++].key);
+                applied_deletes++;
+            }
+
+            // insert the record;
+            pgm.insert_or_assign(insert_vec[i].key, insert_vec[i].value);
+            applied_inserts++;
+        }
+        auto insert_stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(insert_stop - insert_start).count();
+    } 
+
+    progress_update(1.0, "inserting:");
+
+    size_t throughput = (((double) (applied_inserts + applied_deletes) / (double) total_time) * 1e9);
+
+    fprintf(stdout, "%ld\t", throughput);
+}
+
+
+
+static void pgm_rq_bench(PGM &pgm, std::vector<query> queries, size_t trial_cnt=1) 
+{
+    char progbuf[25];
+    sprintf(progbuf, "sampling:");
+
+    size_t batch_size = 100;
+    size_t batches = trial_cnt / batch_size;
+    size_t total_time = 0;
+
+    //std::vector<record> result_set;
+	size_t tot = 0;
+
+    for (int i=0; i<trial_cnt; i++) {
+        auto start = std::chrono::high_resolution_clock::now();
+        for (size_t j=0; j<queries.size(); j++) {
+            auto ptr = pgm.find(queries[j].lower_bound);\
+			tot = 0;
+            while (ptr != pgm.end() && ptr->first <= queries[j].upper_bound) {
+				++tot;
+                //result_set.push_back({ptr->first, ptr->second});
+                ++ptr;
+            }
+			assert(tot > 0);
+            //result_set.clear();
+        }
+        auto stop = std::chrono::high_resolution_clock::now();
+
+        total_time += std::chrono::duration_cast<std::chrono::nanoseconds>(stop - start).count();
+    }
+
+    size_t latency = total_time / (trial_cnt * queries.size());
+
+    fprintf(stdout, "%ld\t", latency);
+}
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: upgm_rq_bench <filename> <record_count> <delete_proportion> <query_file>\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+
+    double insert_batch = 0.5; 
+
+    init_bench_env(record_count, true);
+    auto queries = read_range_queries<query>(qfilename, .0001);
+
+    std::vector<std::pair<key_type, value_type>> data;
+    PGM pgm(data.begin(), data.end());
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<record> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = insert_batch * record_count;
+    warmup(datafile, pgm, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    pgm_rq_insert(pgm, datafile, insert_cnt, delete_prop, to_delete, true);
+    size_t memory_usage = pgm.size_in_bytes();
+    fprintf(stdout, "%ld\t", memory_usage);
+
+    pgm_rq_bench(pgm, queries);
+    fprintf(stdout, "\n");
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}
diff --git a/benchmarks/old-bench/vptree_knn_bench.cpp b/benchmarks/old-bench/vptree_knn_bench.cpp
new file mode 100644
index 0000000..d8247e4
--- /dev/null
+++ b/benchmarks/old-bench/vptree_knn_bench.cpp
@@ -0,0 +1,58 @@
+#include "include/bench.h"
+
+int main(int argc, char **argv)
+{
+    if (argc < 5) {
+        fprintf(stderr, "Usage: vptree_knn_bench <filename> <record_count> <delete_proportion> <query_file> [k]\n");
+        exit(EXIT_FAILURE);
+    }
+
+    std::string filename = std::string(argv[1]);
+    size_t record_count = atol(argv[2]);
+    double delete_prop = atof(argv[3]);
+    std::string qfilename = std::string(argv[4]);
+    size_t k = (argc == 6) ? atol(argv[5]) : 10;
+
+    size_t buffer_cap = 12000;
+    size_t scale_factor = 6;
+    double max_delete_prop = delete_prop;
+
+    init_bench_env(record_count, true);
+    auto queries = read_knn_queries<de::KNNQueryParms<Word2VecRec>>(qfilename, k);
+
+    auto de_vp_knn = ExtendedVPTree_KNN(buffer_cap, scale_factor, max_delete_prop);
+
+    std::fstream datafile;
+    datafile.open(filename, std::ios::in | std::ios::binary);
+
+    std::vector<Word2VecRec> to_delete;
+
+    // warm up the tree with initial_insertions number of initially inserted
+    // records
+    size_t warmup_cnt = 0.1 * record_count;
+    warmup<ExtendedVPTree_KNN, Word2VecRec>(datafile, de_vp_knn, warmup_cnt, delete_prop, to_delete, true, true);
+
+    size_t insert_cnt = record_count - warmup_cnt;
+
+    insert_tput_bench<ExtendedVPTree_KNN, Word2VecRec>(de_vp_knn, datafile, insert_cnt, delete_prop, to_delete, true);
+    fprintf(stdout, "%ld\t", de_vp_knn.get_memory_usage());
+
+    query_latency_bench<ExtendedVPTree_KNN, Word2VecRec, de::KNNQueryParms<Word2VecRec>>(de_vp_knn, queries);
+    fprintf(stdout, "\n");
+
+    auto ts = de_vp_knn.create_static_structure();
+
+    fprintf(stdout, "%ld\t", ts->get_memory_usage());
+    static_latency_bench<de::VPTree<Word2VecRec>, Word2VecRec, de::KNNQueryParms<Word2VecRec>, de::KNNQuery<Word2VecRec>>(
+        ts, queries, 1
+    );
+    fprintf(stdout, "\n");
+
+    delete ts;
+
+    delete_bench_env();
+    fflush(stdout);
+    fflush(stderr);
+
+    exit(EXIT_SUCCESS);
+}