Benchmarks for stream-lib/faster_hll

bench.log

# Measurement Section
# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.avgtime.HLLMergeBenchmark.hllMerge
# Warmup Iteration   1 (3s in 1 thread): 0.377 msec/op
# Warmup Iteration   2 (3s in 1 thread): 0.314 msec/op
# Warmup Iteration   3 (3s in 1 thread): 0.311 msec/op
# Warmup Iteration   4 (3s in 1 thread): 0.311 msec/op
# Warmup Iteration   5 (3s in 1 thread): 0.310 msec/op
Iteration   1 (5s in 1 thread): 0.310 msec/op
Iteration   2 (5s in 1 thread): 0.311 msec/op
Iteration   3 (5s in 1 thread): 0.311 msec/op
Iteration   4 (5s in 1 thread): 0.311 msec/op
Iteration   5 (5s in 1 thread): 0.311 msec/op
Iteration   6 (5s in 1 thread): 0.313 msec/op
Iteration   7 (5s in 1 thread): 0.315 msec/op
Iteration   8 (5s in 1 thread): 0.315 msec/op
Iteration   9 (5s in 1 thread): 0.314 msec/op
Iteration  10 (5s in 1 thread): 0.317 msec/op

Run result "hllMerge": 0.313 ±(95%) 0.002 ±(99%) 0.002 msec/op
Run statistics "hllMerge": min = 0.310, avg = 0.313, max = 0.317, stdev = 0.002
Run confidence intervals "hllMerge": 95% [0.311, 0.314], 99% [0.310, 0.315]


# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.avgtime.HLLMergeBenchmark.hllMergeOrig
# Warmup Iteration   1 (3s in 1 thread): 2.985 msec/op
# Warmup Iteration   2 (3s in 1 thread): 3.360 msec/op
# Warmup Iteration   3 (3s in 1 thread): 2.969 msec/op
# Warmup Iteration   4 (3s in 1 thread): 2.972 msec/op
# Warmup Iteration   5 (3s in 1 thread): 2.969 msec/op
Iteration   1 (5s in 1 thread): 2.965 msec/op
Iteration   2 (5s in 1 thread): 2.966 msec/op
Iteration   3 (5s in 1 thread): 2.969 msec/op
Iteration   4 (5s in 1 thread): 2.956 msec/op
Iteration   5 (5s in 1 thread): 2.957 msec/op
Iteration   6 (5s in 1 thread): 2.962 msec/op
Iteration   7 (5s in 1 thread): 2.963 msec/op
Iteration   8 (5s in 1 thread): 2.961 msec/op
Iteration   9 (5s in 1 thread): 2.959 msec/op
Iteration  10 (5s in 1 thread): 2.960 msec/op

Run result "hllMergeOrig": 2.962 ±(95%) 0.003 ±(99%) 0.004 msec/op
Run statistics "hllMergeOrig": min = 2.956, avg = 2.962, max = 2.969, stdev = 0.004
Run confidence intervals "hllMergeOrig": 95% [2.959, 2.965], 99% [2.958, 2.966]


# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.avgtime.HLLMergeBenchmark.hllPlusMerge
# Warmup Iteration   1 (3s in 1 thread): 0.314 msec/op
# Warmup Iteration   2 (3s in 1 thread): 0.314 msec/op
# Warmup Iteration   3 (3s in 1 thread): 0.299 msec/op
# Warmup Iteration   4 (3s in 1 thread): 0.310 msec/op
# Warmup Iteration   5 (3s in 1 thread): 0.313 msec/op
Iteration   1 (5s in 1 thread): 0.313 msec/op
Iteration   2 (5s in 1 thread): 0.313 msec/op
Iteration   3 (5s in 1 thread): 0.313 msec/op
Iteration   4 (5s in 1 thread): 0.313 msec/op
Iteration   5 (5s in 1 thread): 0.314 msec/op
Iteration   6 (5s in 1 thread): 0.313 msec/op
Iteration   7 (5s in 1 thread): 0.313 msec/op
Iteration   8 (5s in 1 thread): 0.313 msec/op
Iteration   9 (5s in 1 thread): 0.313 msec/op
Iteration  10 (5s in 1 thread): 0.313 msec/op

Run result "hllPlusMerge": 0.313 ±(95%) 0.000 ±(99%) 0.000 msec/op
Run statistics "hllPlusMerge": min = 0.313, avg = 0.313, max = 0.314, stdev = 0.000
Run confidence intervals "hllPlusMerge": 95% [0.313, 0.313], 99% [0.313, 0.314]


# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.avgtime.HLLMergeBenchmark.hllPlusMergeOrig
# Warmup Iteration   1 (3s in 1 thread): 3.116 msec/op
# Warmup Iteration   2 (3s in 1 thread): 3.093 msec/op
# Warmup Iteration   3 (3s in 1 thread): 3.099 msec/op
# Warmup Iteration   4 (3s in 1 thread): 3.102 msec/op
# Warmup Iteration   5 (3s in 1 thread): 3.097 msec/op
Iteration   1 (5s in 1 thread): 3.097 msec/op
Iteration   2 (5s in 1 thread): 3.101 msec/op
Iteration   3 (5s in 1 thread): 3.099 msec/op
Iteration   4 (5s in 1 thread): 3.102 msec/op
Iteration   5 (5s in 1 thread): 3.097 msec/op
Iteration   6 (5s in 1 thread): 3.096 msec/op
Iteration   7 (5s in 1 thread): 3.097 msec/op
Iteration   8 (5s in 1 thread): 3.093 msec/op
Iteration   9 (5s in 1 thread): 3.094 msec/op
Iteration  10 (5s in 1 thread): 3.095 msec/op

Run result "hllPlusMergeOrig": 3.097 ±(95%) 0.002 ±(99%) 0.003 msec/op
Run statistics "hllPlusMergeOrig": min = 3.093, avg = 3.097, max = 3.102, stdev = 0.003
Run confidence intervals "hllPlusMergeOrig": 95% [3.095, 3.099], 99% [3.094, 3.100]


# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.throughput.HLLOfferBenchmark.hllOfferInt
# Warmup Iteration   1 (3s in 1 thread): 86647.912 ops/msec
# Warmup Iteration   2 (3s in 1 thread): 87273.245 ops/msec
# Warmup Iteration   3 (3s in 1 thread): 87920.980 ops/msec
# Warmup Iteration   4 (3s in 1 thread): 87942.991 ops/msec
# Warmup Iteration   5 (3s in 1 thread): 87842.123 ops/msec
Iteration   1 (5s in 1 thread): 87850.120 ops/msec
Iteration   2 (5s in 1 thread): 87852.193 ops/msec
Iteration   3 (5s in 1 thread): 87952.788 ops/msec
Iteration   4 (5s in 1 thread): 88003.762 ops/msec
Iteration   5 (5s in 1 thread): 87756.028 ops/msec
Iteration   6 (5s in 1 thread): 87887.437 ops/msec
Iteration   7 (5s in 1 thread): 88007.154 ops/msec
Iteration   8 (5s in 1 thread): 87904.741 ops/msec
Iteration   9 (5s in 1 thread): 87864.571 ops/msec
Iteration  10 (5s in 1 thread): 87792.039 ops/msec

Run result "hllOfferInt": 87887.083 ±(95%) 59.350 ±(99%) 85.274 ops/msec
Run statistics "hllOfferInt": min = 87756.028, avg = 87887.083, max = 88007.154, stdev = 82.972
Run confidence intervals "hllOfferInt": 95% [87827.733, 87946.434], 99% [87801.810, 87972.357]


# Runtime (per iteration): 5s
# Iterations: 10
# Thread counts (concurrent threads per iteration): [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
# Threads will synchronize iterations
# Running: bench.generated.throughput.HLLOfferBenchmark.hllOfferIntOrig
# Warmup Iteration   1 (3s in 1 thread): 62783.065 ops/msec
# Warmup Iteration   2 (3s in 1 thread): 66192.891 ops/msec
# Warmup Iteration   3 (3s in 1 thread): 65902.016 ops/msec
# Warmup Iteration   4 (3s in 1 thread): 65932.382 ops/msec
# Warmup Iteration   5 (3s in 1 thread): 66018.394 ops/msec
Iteration   1 (5s in 1 thread): 65936.995 ops/msec
Iteration   2 (5s in 1 thread): 65925.273 ops/msec
Iteration   3 (5s in 1 thread): 65999.392 ops/msec
Iteration   4 (5s in 1 thread): 66056.409 ops/msec
Iteration   5 (5s in 1 thread): 66111.274 ops/msec
Iteration   6 (5s in 1 thread): 66028.462 ops/msec
Iteration   7 (5s in 1 thread): 65982.001 ops/msec
Iteration   8 (5s in 1 thread): 66051.908 ops/msec
Iteration   9 (5s in 1 thread): 66003.780 ops/msec
Iteration  10 (5s in 1 thread): 65990.976 ops/msec

Run result "hllOfferIntOrig": 66008.647 ±(95%) 40.107 ±(99%) 57.625 ops/msec
Run statistics "hllOfferIntOrig": min = 65925.273, avg = 66008.647, max = 66111.274, stdev = 56.070
Run confidence intervals "hllOfferIntOrig": 95% [65968.540, 66048.754], 99% [65951.021, 66066.272]


Benchmark                                          Thr    Cnt  Sec         Mean   Mean error          Var    Units
b.g.avgtime.HLLMergeBenchmark.hllMerge               1     10    5        0.313        0.002        0.000  msec/op
b.g.avgtime.HLLMergeBenchmark.hllMergeOrig           1     10    5        2.962        0.004        0.000  msec/op
b.g.avgtime.HLLMergeBenchmark.hllPlusMerge           1     10    5        0.313        0.000        0.000  msec/op
b.g.avgtime.HLLMergeBenchmark.hllPlusMergeOrig       1     10    5        3.097        0.003        0.000  msec/op
b.g.throughput.HLLOfferBenchmark.hllOfferInt         1     10    5    87887.083       85.274     6884.359 ops/msec
b.g.throughput.HLLOfferBenchmark.hllOfferIntOrig     1     10    5    66008.647       57.625     3143.832 ops/msec

HLLMergeBenchmark.java

package bench;

import java.util.Arrays;
import java.util.concurrent.TimeUnit;

import org.openjdk.jmh.annotations.BenchmarkType;
import org.openjdk.jmh.annotations.GenerateMicroBenchmark;
import org.openjdk.jmh.annotations.OutputTimeUnit;
import org.openjdk.jmh.annotations.State;

import com.clearspring.analytics.stream.cardinality.CardinalityMergeException;
import com.clearspring.analytics.stream.cardinality.HyperLogLog;
import com.clearspring.analytics.stream.cardinality.HyperLogLogPlus;
import com.clearspring.analytics.stream.cardinality.ICardinality;

@State
@OutputTimeUnit(TimeUnit.MILLISECONDS)
public class HLLMergeBenchmark {

	final int numToMerge = 5;
	final int bits = 16;
	final int cardinality = 100000;

	HyperLogLog[] hlls = new HyperLogLog[numToMerge];
	HyperLogLogOrig[] hllsOrig = new HyperLogLogOrig[numToMerge];
	
	HyperLogLogPlus[] hllps = new HyperLogLogPlus[numToMerge];
	HyperLogLogPlusOrig[] hllpsOrig = new HyperLogLogPlusOrig[numToMerge];
	
	
	public HLLMergeBenchmark() {
		for (int i = 0; i < numToMerge; i++) {
			hlls[i] = new HyperLogLog(bits);
			hllsOrig[i] = new HyperLogLogOrig(bits);

			hllps[i] = new HyperLogLogPlus(bits);
			hllpsOrig[i] = new HyperLogLogPlusOrig(bits);
			
			for (int j = 0; j < cardinality; j++) {
				double val = Math.random();
				hlls[i].offer(val);
				hllsOrig[i].offer(val);
				hllps[i].offer(val);
				hllpsOrig[i].offer(val);
			}
		}

	}

	@GenerateMicroBenchmark(BenchmarkType.AverageTimePerOp)
	public ICardinality hllMerge() throws CardinalityMergeException {
		HyperLogLog hll = hlls[0];
		HyperLogLog[] tmp = Arrays.copyOfRange(hlls, 1, hlls.length);
		return hll.merge(tmp);
	}

	@GenerateMicroBenchmark(BenchmarkType.AverageTimePerOp)
	public ICardinality hllMergeOrig() throws CardinalityMergeException {
		HyperLogLogOrig hll = hllsOrig[0];
		HyperLogLogOrig[] tmp = Arrays
				.copyOfRange(hllsOrig, 1, hllsOrig.length);
		return hll.merge(tmp);
	}
	
	@GenerateMicroBenchmark(BenchmarkType.AverageTimePerOp)
	public ICardinality hllPlusMerge() throws CardinalityMergeException {
		HyperLogLogPlus hll = hllps[0];
		HyperLogLogPlus[] tmp = Arrays.copyOfRange(hllps, 1, hllps.length);
		return hll.merge(tmp);
	}

	@GenerateMicroBenchmark(BenchmarkType.AverageTimePerOp)
	public ICardinality hllPlusMergeOrig() throws CardinalityMergeException {
		HyperLogLogPlusOrig hll = hllpsOrig[0];
		HyperLogLogPlusOrig[] tmp = Arrays.copyOfRange(hllpsOrig, 1, hllpsOrig.length);
		return hll.merge(tmp);
	}
}

HLLOfferBenchmark.java

package bench;

import java.util.Random;
import java.util.concurrent.TimeUnit;

import org.openjdk.jmh.annotations.BenchmarkType;
import org.openjdk.jmh.annotations.GenerateMicroBenchmark;
import org.openjdk.jmh.annotations.OutputTimeUnit;
import org.openjdk.jmh.annotations.State;

import com.clearspring.analytics.stream.cardinality.HyperLogLog;

@State
@OutputTimeUnit(TimeUnit.MILLISECONDS)
public class HLLOfferBenchmark {

	final Random rand = new Random(42);
	final int bits = 16;

	HyperLogLog hll = new HyperLogLog(bits);
	HyperLogLogOrig hllOrig = new HyperLogLogOrig(bits);

	@GenerateMicroBenchmark(BenchmarkType.OpsPerTimeUnit)
	public void hllOfferInt() {
		hll.offerHashed(rand.nextInt());
	}
	
	@GenerateMicroBenchmark(BenchmarkType.OpsPerTimeUnit)
	public void hllOfferIntOrig() {
		hllOrig.offerHashed(rand.nextInt());
	}
}