package org.jctools.maps.nhbm_test; import java.util.Random; import java.util.concurrent.atomic.AtomicLong; import org.jctools.maps.NonBlockingHashMapLong; /* * Written by Cliff Click and released to the public domain, as explained at * http://creativecommons.org/licenses/publicdomain * Big Chunks of code shamelessly copied from Doug Lea's test harness which is also public domain. */ public class perf_hashlong_test extends Thread { static int _read_ratio, _gr, _pr; static int _thread_min, _thread_max, _thread_incr; static int _table_size; static String KEYS[]; static volatile boolean _start; static volatile boolean _stop; static int check(String arg, String msg, int lower, int upper) throws Exception { return check(Integer.parseInt(arg), msg, lower, upper); } static int check(int x, String msg, int lower, int upper) throws Exception { if (x < lower || x > upper) throw new Error(msg + " must be from " + lower + " to " + upper); return x; } public static void main(String args[]) { if (args.length == 0) { args = new String[] { "75", "2", "8", "2", "100000" }; } // Parse args try { _read_ratio = check(args[0], "read%", 0, 100); _thread_min = check(args[1], "thread_min", 1, 100000); _thread_max = check(args[2], "thread_max", 1, 100000); _thread_incr = check(args[3], "thread_incr", 1, 100000); _table_size = check(args[4], "table_size", 100, 100000000); _gr = (_read_ratio << 20) / 100; _pr = (((1 << 20) - _gr) >> 1) + _gr; int trips = (_thread_max - _thread_min) / _thread_incr; _thread_max = trips * _thread_incr + _thread_min; } catch (Exception e) { System.out.println( "Usage: perf_hash_test read%[0=churn test] thread-min thread-max thread-increment hash_table_size impl[All=0,Hashtable=1,HerlihyHashSet=2,CHM_16=3,CHM_256=4,CHM_4096=5,NonBlockingHashMap=6]"); throw new RuntimeException(e); } System.out.print(_read_ratio + "% gets, " + ((100 - _read_ratio) >> 1) + "% inserts, " + ((100 - _read_ratio) >> 1) + "% removes, " + "table_size=" + _table_size); if (_read_ratio == 0) System.out.print(" -- churn"); String name = "NonBlockingHashMapLong"; System.out.println(" " + name); System.out.println("Threads from " + _thread_min + " to " + _thread_max + " by " + _thread_incr); // Do some warmup int keymax = 1; while (keymax < _table_size) keymax <<= 1; if (_read_ratio == 0) keymax = 1024 * 1024; // The churn test uses a large key set KEYS = new String[keymax]; KEYS[0] = "Cliff0"; for (int i = 1; i < KEYS.length; i++) { KEYS[i] = String.valueOf(i) + "abc" + String.valueOf(i * 17 + 123); } System.out.println("Warmup -variance: "); run_till_stable(Math.min(_thread_min, 2), 1); // Now do the real thing System.out.print("==== Counter Threads Trial: "); int num_trials = 7; // Number of Trials for (int i = 0; i < num_trials; i++) System.out.printf(" %3d ", i); System.out.println(" Avg Stddev"); for (int i = _thread_min; i <= _thread_max; i += _thread_incr) run_till_stable(i, num_trials); } static void run_till_stable(int num_threads, int num_trials) { NonBlockingHashMapLong<String> HM = new NonBlockingHashMapLong<String>(true); String name = "NonBlockingHashMapLong"; System.out.printf("=== %10.10s %3d cnts/sec=", name, num_threads); // Quicky sanity check for (int i = 0; i < 100; i++) { HM.put(i, KEYS[i]); for (int j = 0; j < i; j++) { if (HM.get(j) != KEYS[j]) { throw new Error("Broken table, put " + i + " but cannot find #" + j); } } } long[] trials = new long[num_trials]; // Number of trials long total = 0; for (int j = 0; j < trials.length; j++) { long[] ops = new long[num_threads]; long[] nanos = new long[num_threads]; long millis = run_once(num_threads, HM, ops, nanos); long sum = 0; for (int i = 0; i < num_threads; i++) sum += ops[i]; long ops_per_sec = (sum * 1000L) / millis; trials[j] = ops_per_sec; total += ops_per_sec; System.out.printf(" %10d", ops_per_sec); // for( int i=0; i<num_threads; i++ ) { // if( nanos[i] < 1980000000 || // nanos[i] > 2010000000 || // ops[i] < 100000 ) // System.out.printf(" %d",ops[i]); // } } if (trials.length > 2) { // Toss out low & high int lo = 0; int hi = 0; for (int j = 1; j < trials.length; j++) { if (trials[lo] < trials[j]) lo = j; if (trials[hi] > trials[j]) hi = j; } total -= (trials[lo] + trials[hi]); trials[lo] = trials[trials.length - 1]; trials[hi] = trials[trials.length - 2]; // Print avg,stddev long avg = total / (trials.length - 2); long stddev = compute_stddev(trials, trials.length - 2); long p = stddev * 100 / avg; // std-dev as a percent if (trials.length - 2 > 2) { // Toss out low & high lo = 0; hi = 0; for (int j = 1; j < trials.length - 2; j++) { if (trials[lo] < trials[j]) lo = j; if (trials[hi] > trials[j]) hi = j; } total -= (trials[lo] + trials[hi]); trials[lo] = trials[trials.length - 2 - 1]; trials[hi] = trials[trials.length - 2 - 2]; // Print avg,stddev avg = total / (trials.length - 2 - 2); stddev = compute_stddev(trials, trials.length - 2 - 2); p = stddev * 100 / avg; // std-dev as a percent } System.out.printf(" %10d", avg); System.out.printf(" (+/-%2d%%) %d", p, HM.size()); } System.out.println(); } static long compute_stddev(long[] trials, int len) { double sum = 0; double squ = 0.0; for (int i = 0; i < len; i++) { double d = (double) trials[i]; sum += d; squ += d * d; } double x = squ - sum * sum / len; double stddev = Math.sqrt(x / (len - 1)); return (long) stddev; } // Worker thread fields final int _tnum; final NonBlockingHashMapLong<String> _hash; // Shared hashtable final long[] _ops; final long[] _nanos; public perf_hashlong_test() { _tnum = 0; _hash = null; _ops = null; _nanos = null; } private perf_hashlong_test(int tnum, NonBlockingHashMapLong<String> HM, long[] ops, long[] nanos) { _tnum = tnum; _hash = HM; _ops = ops; _nanos = nanos; } static long run_once(int num_threads, NonBlockingHashMapLong<String> HM, long[] ops, long[] nanos) { Random R = new Random(); _start = false; _stop = false; HM.put(0, "Cliff0"); HM.remove(0); int sz = HM.size(); while (sz + 1024 < _table_size) { int idx = R.nextInt(); for (int i = 0; i < 1024; i++) { int k = idx & (KEYS.length - 1); HM.put(k, KEYS[k]); idx++; } sz = HM.size(); } while (sz < ((_table_size >> 1) + (_table_size >> 3))) { int trip = 0; int idx = R.nextInt(); while (true) { int k = idx & (KEYS.length - 1); String key = KEYS[k]; if (sz < _table_size) { if (HM.put(k, key) == null) { sz++; break; } } else { if (HM.remove(k) != null) { sz--; break; } } idx++; if ((trip & 15) == 15) idx = R.nextInt(); if (trip++ > 1024 * 1024) { if (trip > 1024 * 1024 + 100) throw new RuntimeException( "barf trip " + sz + " " + HM.size() + " numkeys=" + KEYS.length); System.out.println(key); } } } if (sz != HM.size()) { throw new Error("size does not match table contents sz=" + sz + " size()=" + HM.size()); } // Launch threads // long nanoz = System.nanoTime(); // System.out.println(" "+nanoz+" Create-Threads"); perf_hashlong_test thrs[] = new perf_hashlong_test[num_threads]; for (int i = 0; i < num_threads; i++) thrs[i] = new perf_hashlong_test(i, HM, ops, nanos); for (int i = 0; i < num_threads; i++) thrs[i].start(); // Run threads // long nano = System.nanoTime(); // System.out.println(" "+nano+" Start"); long start = System.currentTimeMillis(); _start = true; try { Thread.sleep(2000); } catch (InterruptedException e) { /* empty */} _stop = true; long stop = System.currentTimeMillis(); // long nanox = System.nanoTime(); long millis = stop - start; // System.out.println(" "+nanox+" Stop"); for (int i = 0; i < num_threads; i++) try { thrs[i].join(); } catch (InterruptedException ie) { throw new RuntimeException(ie); } // long nanoy = System.nanoTime(); // System.out.println(" "+nanoy+" Join-Done"); return millis; } // What a worker thread does public void run() { if (_read_ratio == 0) { run_churn(); } else { run_normal(); } } // Force a large turnover of live keys, while keeping the total live-set // low. 10 keys kept alive per thread, out of a set of a million or so. // constantly churned, so we constantly need to 'cleanse' the table to flush // old entries. public void run_churn() { int reprobe = System.identityHashCode(Thread.currentThread()); int idx = reprobe; while (!_start) // Spin till Time To Go try { Thread.sleep(1); } catch (Exception e) { /* empty */} long nano1 = System.nanoTime(); int get_ops = 0; int put_ops = 0; int del_ops = 0; while (!_stop) { // Insert a key 10 probes in the future, // remove a key 0 probes in the future, // Net result is the thread keeps 10 random keys in table int k1 = (idx + reprobe * 10) & (KEYS.length - 1); String key1 = KEYS[k1]; _hash.put(k1, key1); put_ops++; // Remove a key 0 probes in the future int k2 = idx & (KEYS.length - 1); _hash.remove(k2); del_ops++; idx += reprobe; } // We stopped; report results into shared result structure long nano2 = System.nanoTime(); int total = get_ops + put_ops + del_ops; _ops[_tnum] = total; _nanos[_tnum] = (nano2 - nano1); } public void run_normal() { SimpleRandom R = new SimpleRandom(); while (!_start) // Spin till Time To Go try { Thread.sleep(1); } catch (Exception e) { /* empty */} long nano1 = System.nanoTime(); int get_ops = 0; int put_ops = 0; int del_ops = 0; while (!_stop) { int x = R.nextInt() & ((1 << 20) - 1); int k = R.nextInt() & (KEYS.length - 1); String key = KEYS[k]; if (x < _gr) { get_ops++; String val = _hash.get(k); if (val != null && !val.equals(key)) throw new IllegalArgumentException("Mismatched key=" + key + " and val=" + val); } else if (x < _pr) { put_ops++; _hash.putIfAbsent(k, key); // An interesting version: testing get immediately after putIfAbsent. // Of course in a multi-threaded context it immediately throws false-positives. // if( _hash.putIfAbsent( key, key ) == null ) // if( _hash.get(key) == null ) // throw new Error("putIfAbsent failed to put key=" + key + " for put_ops=" + put_ops + "and // getops=" +get_ops + " del_ops="+del_ops); } else { del_ops++; _hash.remove(k); } } // We stopped; report results into shared result structure long nano2 = System.nanoTime(); int total = get_ops + put_ops + del_ops; _ops[_tnum] = total; _nanos[_tnum] = (nano2 - nano1); } // Fairly fast random numbers public static final class SimpleRandom { private final static long multiplier = 0x5DEECE66DL; private final static long addend = 0xBL; private final static long mask = (1L << 48) - 1; static final AtomicLong seq = new AtomicLong(-715159705); private long seed; SimpleRandom() { seed = System.nanoTime() + seq.getAndAdd(129); } public int nextInt() { return next(); } public int next() { long nextseed = (seed * multiplier + addend) & mask; seed = nextseed; return ((int) (nextseed >>> 17)) & 0x7FFFFFFF; } } }