1brc/src/main/java/dev/morling/onebrc/CalculateAverage_phd3.java
Pratham 6c7012a43e
Add improvements ()
- custom hashmap
- avoid string creation
- use graal
2024-01-15 18:47:06 +01:00

340 lines
13 KiB
Java

/*
* Copyright 2023 The original authors
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package dev.morling.onebrc;
import static java.util.stream.Collectors.*;
import java.io.File;
import java.io.RandomAccessFile;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.TreeMap;
import java.util.concurrent.Callable;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.stream.Collectors;
import java.util.stream.IntStream;
public class CalculateAverage_phd3 {
private static final int NUM_THREADS = Runtime.getRuntime().availableProcessors() * 2;
private static final String FILE = "./measurements.txt";
private static final long FILE_SIZE = new File(FILE).length();
// A chunk is a unit for processing, the file will be divided in chunks of the following size
private static final int CHUNK_SIZE = 65536 * 1024;
// Read a little more data into the buffer to finish processing current line
private static final int PADDING = 512;
// Minor : Precompute powers to avoid recalculating while parsing doubles (temperatures)
private static final double[] POWERS_OF_10 = IntStream.range(0, 6).mapToDouble(x -> Math.pow(10.0, x)).toArray();
/**
* A Utility to print aggregated information in the desired format
*/
private record ResultRow(double min, double mean, double max) {
public String toString() {
return round(min) + "/" + round(mean) + "/" + round(max);
}
private double round(double value) {
return Math.round(value * 10.0) / 10.0;
}
};
public static ResultRow resultRow(AggregationInfo aggregationInfo) {
return new ResultRow(aggregationInfo.min, (Math.round(aggregationInfo.sum * 10.0) / 10.0) / (aggregationInfo.count), aggregationInfo.max);
}
public static void main(String[] args) throws Exception {
long fileLength = new File(FILE).length();
int numChunks = (int) Math.ceil(fileLength * 1.0 / CHUNK_SIZE);
ExecutorService executorService = Executors.newFixedThreadPool(NUM_THREADS);
BufferDataProvider provider = new RandomAccessBasedProvider(FILE, FILE_SIZE);
List<Future<LinearProbingHashMap>> futures = new ArrayList<>();
// Process chunks in parallel
for (int chunkIndex = 0; chunkIndex < numChunks; chunkIndex++) {
futures.add(executorService.submit(new Aggregator(chunkIndex, provider)));
}
executorService.shutdown();
executorService.awaitTermination(10, TimeUnit.MINUTES);
Map<String, AggregationInfo> info = futures.stream().map(f -> {
try {
return f.get();
}
catch (ExecutionException | InterruptedException e) {
throw new RuntimeException(e);
}
})
.map(LinearProbingHashMap::toMap)
.flatMap(map -> map.entrySet().stream())
.sequential()
.collect(Collectors.toMap(Map.Entry::getKey, Map.Entry::getValue, AggregationInfo::update));
Map<String, ResultRow> measurements = new TreeMap<>(info.entrySet().stream()
.collect(toMap(Map.Entry::getKey, e -> resultRow(e.getValue()))));
System.out.println(measurements);
}
/**
* Stores required running aggregation information to be able to compute min/max/average at the end
*/
private static class AggregationInfo {
double min = Double.POSITIVE_INFINITY;
double max = Double.NEGATIVE_INFINITY;
double sum;
long count;
public AggregationInfo update(AggregationInfo update) {
this.count += update.count;
this.sum += update.sum;
if (this.max < update.max) {
this.max = update.max;
}
if (this.min > update.min) {
this.min = update.min;
}
return this;
}
public AggregationInfo update(double value) {
this.count++;
this.sum += value;
if (this.max < value) {
this.max = value;
}
if (this.min > value) {
this.min = value;
}
return this;
}
}
private interface BufferDataProvider {
int read(byte[] buffer, long offset) throws Exception;
}
/**
* uses RandomAccessFile seek and read APIs to load data into a buffer.
*/
private static class RandomAccessBasedProvider implements BufferDataProvider {
private final String filePath;
RandomAccessBasedProvider(String filePath, long fileSize) {
this.filePath = filePath;
}
@Override
public int read(byte[] buffer, long offset) throws Exception {
RandomAccessFile file = null;
try {
file = new RandomAccessFile(filePath, "r");
file.seek(offset);
return file.read(buffer);
}
finally {
if (file != null) {
file.close();
}
}
}
}
/**
* Task to processes a chunk of file and return a custom linear probing hashmap for performance
*/
private static class Aggregator implements Callable<LinearProbingHashMap> {
private final long startByte;
private final BufferDataProvider dataProvider;
public Aggregator(long chunkIndex, BufferDataProvider dataProvider) {
this.startByte = chunkIndex * CHUNK_SIZE;
this.dataProvider = dataProvider;
}
@Override
public LinearProbingHashMap call() {
try {
// offset for the last byte to be processed (excluded)
long endByte = Math.min(startByte + CHUNK_SIZE, FILE_SIZE);
// read a little more than needed to cover next entry if needed
long bufferSize = endByte - startByte + ((endByte == FILE_SIZE) ? 0 : PADDING);
byte[] buffer = new byte[(int) bufferSize];
int bytes = dataProvider.read(buffer, startByte);
// Partial aggregation in a hashmap
return processBuffer(buffer, startByte == 0, endByte - startByte);
}
catch (Throwable e) {
throw new RuntimeException(e);
}
}
private static LinearProbingHashMap processBuffer(byte[] buffer, boolean isFileStart, long nextChunkStart) {
int start = 0;
// Move to the next entry after '\n'. Don't do this if we're at the start of
// the file to avoid missing first entry.
if (!isFileStart) {
while (buffer[start] != '\n') {
start++;
}
start += 1;
}
LinearProbingHashMap chunkLocalMap = new LinearProbingHashMap();
while (true) {
LineInfo lineInfo = getNextLine(buffer, start);
byte[] keyBytes = new byte[lineInfo.semicolonIndex - start];
System.arraycopy(buffer, start, keyBytes, 0, keyBytes.length);
double value = parseDouble(buffer, lineInfo.semicolonIndex + 1, lineInfo.nextStart - 1);
// Update aggregated value for the given key with the new line
AggregationInfo info = chunkLocalMap.get(keyBytes, lineInfo.keyHash);
info.update(value);
if ((lineInfo.nextStart > nextChunkStart) || (lineInfo.nextStart >= buffer.length)) {
// we are already at a point where the next line will be processed in the next chunk,
// so the job is done here
break;
}
start = lineInfo.nextStart();
}
return chunkLocalMap;
}
/**
* Converts bytes to double value without intermediate string conversion, faster than Double.parseDouble.
*/
private static double parseDouble(byte[] bytes, int offset, int end) {
boolean negative = (bytes[offset] == '-');
int current = negative ? offset + 1 : offset;
int preFloat = 0;
while (current < end && bytes[current] != '.') {
preFloat = (preFloat * 10) + (bytes[current++] - '0');
}
current++;
int postFloatStart = current;
int postFloat = 0;
while (current < end) {
postFloat = (postFloat * 10) + (bytes[current++] - '0');
}
return (preFloat + ((postFloat) / POWERS_OF_10[end - postFloatStart])) * (negative ? -1 : 1);
}
/**
* Identifies indexes of the next ';' and '\n', which will be used to get entry key and value from line. Also
* computes the hash value for the key while iterating.
*/
private static LineInfo getNextLine(byte[] buffer, int start) {
// caller guarantees that the access is in bounds, so no index check
int hash = 0;
while (buffer[start] != ';') {
start++;
hash = hash * 31 + buffer[start];
}
// The following is just to further reduce the probability of collisions
hash = hash ^ (hash << 16);
int semicolonIndex = start;
// caller guarantees that the access is in bounds, so no index check
while (buffer[start] != '\n') {
start++;
}
return new LineInfo(semicolonIndex, start + 1, hash);
}
}
private record LineInfo(int semicolonIndex, int nextStart, int keyHash) {
}
/**
* A simple map with pre-configured fixed bucket count. With 2^13 buckets and current hash function, seeing 4
* collisions which is not too bad. Every bucket is implemented with a linked list. The map is NOT thread safe.
*/
private static class LinearProbingHashMap {
private final static int BUCKET_COUNT = 8191;
private final Node[] buckets;
LinearProbingHashMap() {
this.buckets = new Node[BUCKET_COUNT];
}
/**
* Given a key, returns the current value of AggregationInfo. If not present, creates a new empty node at the
* front of the bucket
*/
public AggregationInfo get(byte[] key, int keyHash) {
// find bucket index through bitwise AND, works for bucketCount = (2^p - 1)
int bucketIndex = BUCKET_COUNT & keyHash;
Node current = buckets[bucketIndex];
while (current != null) {
if (Arrays.equals(current.entry.key(), key)) {
return current.entry.aggregationInfo();
}
current = current.next;
}
// Entry does not exist, so add a new node in the linked list
AggregationInfo newInfo = new AggregationInfo();
KeyValuePair pair = new KeyValuePair(key, keyHash, newInfo);
Node newNode = new Node(pair, buckets[bucketIndex]);
buckets[bucketIndex] = newNode;
return newNode.entry.aggregationInfo();
}
/**
* A helper to convert to Java's hash map to build the final aggregation after partial aggregations
*/
private Map<String, AggregationInfo> toMap() {
Map<String, AggregationInfo> map = new HashMap<>();
for (Node bucket : buckets) {
while (bucket != null) {
map.put(new String(bucket.entry.key, StandardCharsets.UTF_8), bucket.entry.aggregationInfo());
bucket = bucket.next;
}
}
return map;
}
}
/**
* Linked List node to implement a bucket of custom hash map
*/
private static class Node {
KeyValuePair entry;
Node next;
public Node(KeyValuePair entry, Node next) {
this.entry = entry;
this.next = next;
}
}
/**
* a wrapper class to store information needed for storing a measurement information in the hashmap
*/
private record KeyValuePair(byte[] key, int keyHash, AggregationInfo aggregationInfo) {
}
}