seaweedfs/go/weed/benchmark.go

381 lines
10 KiB
Go
Raw Normal View History

2014-03-10 02:42:50 +00:00
package main
import (
"bufio"
"code.google.com/p/weed-fs/go/glog"
"code.google.com/p/weed-fs/go/operation"
"code.google.com/p/weed-fs/go/util"
"fmt"
"io"
2014-03-10 06:12:05 +00:00
"math"
"math/rand"
2014-03-10 02:42:50 +00:00
"os"
"runtime"
2014-03-10 02:42:50 +00:00
"strings"
"sync"
"time"
)
type BenchmarkOptions struct {
server *string
concurrency *int
numberOfFiles *int
fileSize *int
idListFile *string
write *bool
read *bool
sequentialRead *bool
collection *string
vid2server map[string]string //cache for vid locations
2014-03-10 02:42:50 +00:00
}
var (
b BenchmarkOptions
)
func init() {
cmdBenchmark.Run = runbenchmark // break init cycle
cmdBenchmark.IsDebug = cmdBenchmark.Flag.Bool("debug", false, "verbose debug information")
b.server = cmdBenchmark.Flag.String("server", "localhost:9333", "weedfs master location")
b.concurrency = cmdBenchmark.Flag.Int("c", 7, "number of concurrent write or read processes")
b.fileSize = cmdBenchmark.Flag.Int("size", 1024, "simulated file size in bytes")
b.numberOfFiles = cmdBenchmark.Flag.Int("n", 1024*1024, "number of files to write for each thread")
b.idListFile = cmdBenchmark.Flag.String("list", os.TempDir()+"/benchmark_list.txt", "list of uploaded file ids")
b.write = cmdBenchmark.Flag.Bool("write", true, "enable write")
b.read = cmdBenchmark.Flag.Bool("read", true, "enable read")
b.sequentialRead = cmdBenchmark.Flag.Bool("readSequentially", false, "randomly read by ids from \"-list\" specified file")
b.collection = cmdBenchmark.Flag.String("collection", "benchmark", "write data to this collection")
2014-03-10 02:42:50 +00:00
}
var cmdBenchmark = &Command{
UsageLine: "benchmark -server=localhost:9333 -c=10 -n=100000",
Short: "benchmark on writing millions of files and read out",
Long: `benchmark on an empty weed file system.
Two tests during benchmark:
1) write lots of small files to the system
2) read the files out
The file content is mostly zero, but no compression is done.
By default, write 1 million files of 1KB each with 7 concurrent threads,
and randomly read them out with 7 concurrent threads.
You can choose to only benchmark read or write.
During write, the list of uploaded file ids is stored in "-list" specified file.
You can also use your own list of file ids to run read test.
Write speed and read speed will be collected.
The numbers are used to get a sense of the system.
But usually your network or the hard drive is
the real bottleneck.
`,
}
var (
2014-03-10 06:12:05 +00:00
wait sync.WaitGroup
writeStats *stats
readStats *stats
2014-03-10 02:42:50 +00:00
)
func runbenchmark(cmd *Command, args []string) bool {
finishChan := make(chan bool)
fileIdLineChan := make(chan string)
b.vid2server = make(map[string]string)
fmt.Printf("This is Weed File System version %s %s %s\n", VERSION, runtime.GOOS, runtime.GOARCH)
2014-03-10 02:42:50 +00:00
if *b.write {
writeStats = newStats()
idChan := make(chan int)
wait.Add(*b.concurrency)
go writeFileIds(*b.idListFile, fileIdLineChan, finishChan)
for i := 0; i < *b.concurrency; i++ {
2014-03-10 06:12:05 +00:00
go writeFiles(idChan, fileIdLineChan, writeStats)
2014-03-10 02:42:50 +00:00
}
2014-03-10 06:12:05 +00:00
writeStats.start = time.Now()
go writeStats.checkProgress("Writing Benchmark", finishChan)
2014-03-10 02:42:50 +00:00
for i := 0; i < *b.numberOfFiles; i++ {
idChan <- i
}
close(idChan)
wait.Wait()
2014-03-10 06:12:05 +00:00
writeStats.end = time.Now()
2014-03-10 02:42:50 +00:00
wait.Add(1)
finishChan <- true
finishChan <- true
2014-03-10 02:42:50 +00:00
wait.Wait()
writeStats.printStats()
2014-03-10 02:42:50 +00:00
}
if *b.read {
readStats = newStats()
wait.Add(*b.concurrency)
go readFileIds(*b.idListFile, fileIdLineChan)
2014-03-10 06:12:05 +00:00
readStats.start = time.Now()
go readStats.checkProgress("Randomly Reading Benchmark", finishChan)
2014-03-10 02:42:50 +00:00
for i := 0; i < *b.concurrency; i++ {
2014-03-10 06:12:05 +00:00
go readFiles(fileIdLineChan, readStats)
2014-03-10 02:42:50 +00:00
}
wait.Wait()
finishChan <- true
2014-03-10 06:12:05 +00:00
readStats.end = time.Now()
readStats.printStats()
2014-03-10 02:42:50 +00:00
}
return true
}
2014-03-10 06:12:05 +00:00
func writeFiles(idChan chan int, fileIdLineChan chan string, s *stats) {
2014-03-10 02:42:50 +00:00
for {
if id, ok := <-idChan; ok {
start := time.Now()
fp := &operation.FilePart{Reader: &FakeReader{id: uint64(id), size: int64(*b.fileSize)}, FileSize: int64(*b.fileSize)}
if assignResult, err := operation.Assign(*b.server, 1, "", *b.collection); err == nil {
fp.Server, fp.Fid, fp.Collection = assignResult.PublicUrl, assignResult.Fid, *b.collection
fp.Upload(0, *b.server)
2014-03-10 02:42:50 +00:00
writeStats.addSample(time.Now().Sub(start))
fileIdLineChan <- fp.Fid
2014-03-10 06:12:05 +00:00
s.transferred += int64(*b.fileSize)
s.completed++
2014-03-10 02:42:50 +00:00
if *cmdBenchmark.IsDebug {
fmt.Printf("writing %d file %s\n", id, fp.Fid)
}
} else {
2014-03-10 06:12:05 +00:00
s.failed++
2014-03-10 02:42:50 +00:00
println("writing file error:", err.Error())
}
} else {
break
}
}
wait.Done()
}
2014-03-10 06:12:05 +00:00
func readFiles(fileIdLineChan chan string, s *stats) {
2014-03-10 02:42:50 +00:00
for {
if fid, ok := <-fileIdLineChan; ok {
if len(fid) == 0 {
continue
}
if fid[0] == '#' {
continue
}
if *cmdBenchmark.IsDebug {
fmt.Printf("reading file %s\n", fid)
}
parts := strings.SplitN(fid, ",", 2)
vid := parts[0]
start := time.Now()
if server, ok := b.vid2server[vid]; !ok {
if ret, err := operation.Lookup(*b.server, vid); err == nil {
if len(ret.Locations) > 0 {
server = ret.Locations[0].PublicUrl
b.vid2server[vid] = server
}
}
}
if server, ok := b.vid2server[vid]; ok {
url := "http://" + server + "/" + fid
if bytesRead, err := util.Get(url); err == nil {
2014-03-10 06:12:05 +00:00
s.completed++
s.transferred += int64(len(bytesRead))
2014-03-10 02:42:50 +00:00
readStats.addSample(time.Now().Sub(start))
} else {
2014-03-10 06:12:05 +00:00
s.failed++
2014-03-10 02:42:50 +00:00
println("!!!! Failed to read from ", url, " !!!!!")
}
} else {
2014-03-10 06:12:05 +00:00
s.failed++
2014-03-10 02:42:50 +00:00
println("!!!! volume id ", vid, " location not found!!!!!")
}
} else {
break
}
}
wait.Done()
}
func writeFileIds(fileName string, fileIdLineChan chan string, finishChan chan bool) {
file, err := os.OpenFile(fileName, os.O_WRONLY|os.O_CREATE|os.O_TRUNC, 0644)
if err != nil {
glog.Fatalf("File to create file %s: %s\n", fileName, err)
}
defer file.Close()
for {
select {
case <-finishChan:
wait.Done()
return
case line := <-fileIdLineChan:
file.Write([]byte(line))
file.Write([]byte("\n"))
}
}
}
func readFileIds(fileName string, fileIdLineChan chan string) {
file, err := os.Open(fileName) // For read access.
if err != nil {
glog.Fatalf("File to read file %s: %s\n", fileName, err)
}
defer file.Close()
r := bufio.NewReader(file)
if *b.sequentialRead {
for {
if line, err := Readln(r); err == nil {
fileIdLineChan <- string(line)
} else {
break
}
}
} else {
lines := make([]string, 0, *b.numberOfFiles)
for {
if line, err := Readln(r); err == nil {
lines = append(lines, string(line))
} else {
break
}
}
for i := 0; i < *b.numberOfFiles; i++ {
fileIdLineChan <- lines[rand.Intn(len(lines))]
2014-03-10 02:42:50 +00:00
}
}
2014-03-10 02:42:50 +00:00
close(fileIdLineChan)
}
2014-03-10 06:12:05 +00:00
const (
benchResolution = 10000 //0.1 microsecond
benchBucket = 1000000000 / benchResolution
)
2014-03-10 02:42:50 +00:00
type stats struct {
2014-03-10 06:12:05 +00:00
data []int
completed int
failed int
transferred int64
start time.Time
end time.Time
2014-03-10 02:42:50 +00:00
}
2014-03-10 06:12:05 +00:00
var percentages = []int{50, 66, 75, 80, 90, 95, 98, 99, 100}
func newStats() *stats {
return &stats{data: make([]int, benchResolution)}
2014-03-10 02:42:50 +00:00
}
func (s *stats) addSample(d time.Duration) {
2014-03-10 06:12:05 +00:00
s.data[int(d/benchBucket)]++
2014-03-10 02:42:50 +00:00
}
2014-03-10 06:12:05 +00:00
func (s *stats) checkProgress(testName string, finishChan chan bool) {
2014-03-10 06:12:05 +00:00
fmt.Printf("\n------------ %s ----------\n", testName)
ticker := time.Tick(time.Second)
for {
select {
case <-finishChan:
break
case <-ticker:
fmt.Printf("Completed %d of %d requests, %3.1f%%\n", s.completed, *b.numberOfFiles, float64(s.completed)*100/float64(*b.numberOfFiles))
}
}
}
func (s *stats) printStats() {
2014-03-10 06:12:05 +00:00
timeTaken := float64(int64(s.end.Sub(s.start))) / 1000000000
2014-03-10 20:22:08 +00:00
fmt.Printf("\nConcurrency Level: %d\n", *b.concurrency)
2014-03-10 06:12:05 +00:00
fmt.Printf("Time taken for tests: %.3f seconds\n", timeTaken)
fmt.Printf("Complete requests: %d\n", s.completed)
fmt.Printf("Failed requests: %d\n", s.failed)
fmt.Printf("Total transferred: %d bytes\n", s.transferred)
fmt.Printf("Requests per second: %.2f [#/sec]\n", float64(s.completed)/timeTaken)
fmt.Printf("Transfer rate: %.2f [Kbytes/sec]\n", float64(s.transferred)/1024/timeTaken)
n, sum := 0, 0
min, max := 10000000, 0
for i := 0; i < len(s.data); i++ {
n += s.data[i]
sum += s.data[i] * i
if s.data[i] > 0 {
if min > i {
min = i
}
if max < i {
max = i
}
}
}
avg := float64(sum) / float64(n)
varianceSum := 0.0
for i := 0; i < len(s.data); i++ {
if s.data[i] > 0 {
d := float64(i) - avg
varianceSum += d * d * float64(s.data[i])
}
}
std := math.Sqrt(varianceSum / float64(n))
fmt.Printf("\nConnection Times (ms)\n")
fmt.Printf(" min avg max std\n")
fmt.Printf("Total: %2.1f %3.1f %3.1f %3.1f\n", float32(min)/10, float32(avg)/10, float32(max)/10, std/10)
//printing percentiles
fmt.Printf("\nPercentage of the requests served within a certain time (ms)\n")
percentiles := make([]int, len(percentages))
for i := 0; i < len(percentages); i++ {
percentiles[i] = n * percentages[i] / 100
}
percentiles[len(percentiles)-1] = n
percentileIndex := 0
currentSum := 0
for i := 0; i < len(s.data); i++ {
currentSum += s.data[i]
if s.data[i] > 0 && percentileIndex < len(percentiles) && currentSum >= percentiles[percentileIndex] {
fmt.Printf(" %3d%% %5.1f ms\n", percentages[percentileIndex], float32(i)/10.0)
percentileIndex++
for percentileIndex < len(percentiles) && currentSum >= percentiles[percentileIndex] {
percentileIndex++
}
2014-03-10 02:42:50 +00:00
}
}
}
// a fake reader to generate content to upload
type FakeReader struct {
id uint64 // an id number
size int64 // max bytes
}
func (l *FakeReader) Read(p []byte) (n int, err error) {
if l.size <= 0 {
return 0, io.EOF
}
if int64(len(p)) > l.size {
n = int(l.size)
} else {
n = len(p)
}
for i := 0; i < n-8; i += 8 {
for s := uint(0); s < 8; s++ {
p[i] = byte(l.id >> (s * 8))
}
}
l.size -= int64(n)
return
}
func Readln(r *bufio.Reader) ([]byte, error) {
var (
isPrefix bool = true
err error = nil
line, ln []byte
)
for isPrefix && err == nil {
line, isPrefix, err = r.ReadLine()
ln = append(ln, line...)
}
return ln, err
}