example/benchmark-read-throughput: measure read throughput

This helps settle how much performance improvement splice()ing files
brings. The answer is that end-to-end read performance is improved by
about 10% for reading large files.

example/benchmark-read-throughput/readbench.go

+// Copyright 2017 the Go-FUSE Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// readbench is a benchmark helper for measuring throughput on
+// single-file reads out of a FUSE filesystem.
+package main
+
+import (
+	"flag"
+	"fmt"
+	"log"
+	"os"
+	"time"
+)
+
+func gulp(fn string, bs int) (int, error) {
+	f, err := os.Open(fn)
+	if err != nil {
+		return 0, err
+	}
+	defer f.Close()
+
+	var tot int
+	buf := make([]byte, bs)
+	for {
+		n, _ := f.Read(buf[:])
+		tot += n
+		if n < len(buf) {
+			break
+		}
+	}
+
+	return tot, nil
+}
+
+func main() {
+	bs := flag.Int("bs", 32, "blocksize in kb")
+	mbLimit := flag.Int("limit", 1000, "amount of data to read in mb")
+	flag.Parse()
+	if len(flag.Args()) < 1 {
+		log.Fatal("readbench [-bs BLOCKSIZE -limit SIZE] file")
+	}
+	blocksize := *bs * 1024
+	totMB := 0.0
+	var totDT time.Duration
+
+	for totMB < float64(*mbLimit) {
+		t := time.Now()
+		n, err := gulp(flag.Arg(0), blocksize)
+		if err != nil {
+			log.Fatal(err)
+		}
+		dt := time.Now().Sub(t)
+		mb := float64(n) / (1 << 20)
+
+		totMB += mb
+		totDT += dt
+	}
+	fmt.Printf("block size %d kb: %.1f MB in %v: %.2f MBs/s\n", *bs, totMB, totDT, totMB/float64(totDT)*1e9)
+}