diff --git a/s2/cmd/internal/readahead/LICENSE b/s2/cmd/internal/readahead/LICENSE
new file mode 100644
index 0000000000..eaeb61a87e
--- /dev/null
+++ b/s2/cmd/internal/readahead/LICENSE
@@ -0,0 +1,22 @@
+The MIT License (MIT)
+
+Copyright (c) 2015 Klaus Post
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
diff --git a/s2/cmd/internal/readahead/README.md b/s2/cmd/internal/readahead/README.md
new file mode 100644
index 0000000000..e6d0e92362
--- /dev/null
+++ b/s2/cmd/internal/readahead/README.md
@@ -0,0 +1,57 @@
+# readahead
+Asynchronous read-ahead for Go readers
+
+This package will allow you to add readhead to any reader. This means a separate goroutine will perform reads from your upstream reader, so you can request from this reader without delay.
+
+This is helpful for splitting an input stream into concurrent processing, and also helps smooth out **bursts** of input or output.
+
+This should be fully transparent, except that once an error has been returned from the Reader, it will not recover. A panic will be caught and returned as an error.
+
+The readahead object also fulfills the [`io.WriterTo`](https://golang.org/pkg/io/#WriterTo) interface, which is likely to speed up `io.Copy` and other code that use the interface.
+
+See an introduction: [An Async Read-ahead Package for Go](https://blog.klauspost.com/an-async-read-ahead-package-for-go/)
+
+[![GoDoc][1]][2] [![Build Status][3]][4]
+
+[1]: https://godoc.org/github.com/klauspost/readahead?status.svg
+[2]: https://godoc.org/github.com/klauspost/readahead
+[3]: https://travis-ci.org/klauspost/readahead.svg
+[4]: https://travis-ci.org/klauspost/readahead
+
+# usage
+
+To get the package use `go get -u github.com/klauspost/readahead`.
+
+Here is a simple example that does file copy. Error handling has been omitted for brevity.
+```Go
+input, _ := os.Open("input.txt")
+output, _ := os.Create("output.txt")
+defer input.Close()
+defer output.Close()
+
+// Create a read-ahead Reader with default settings
+ra := readahead.NewReader(input)
+defer ra.Close()
+
+// Copy the content to our output
+_, _ = io.Copy(output, ra)
+```
+
+# settings
+
+You can finetune the read-ahead for your specific use case, and adjust the number of buffers and the size of each buffer.
+
+The default the size of each buffer is 1MB, and there are 4 buffers. Do not make your buffers too small since there is a small overhead for passing buffers between goroutines. Other than that you are free to experiment with buffer sizes.
+
+# contributions
+
+On this project contributions in terms of new features is limited to:
+
+* Features that are widely usable and
+* Features that have extensive tests
+
+This package is meant to be simple and stable, so therefore these strict requirements.
+
+# license
+
+This package is released under the MIT license. See the supplied LICENSE file for more info.
diff --git a/s2/cmd/internal/readahead/reader.go b/s2/cmd/internal/readahead/reader.go
new file mode 100644
index 0000000000..168932cd4f
--- /dev/null
+++ b/s2/cmd/internal/readahead/reader.go
@@ -0,0 +1,275 @@
+// Copyright (c) 2015 Klaus Post, released under MIT License. See LICENSE file.
+
+// The readahead package will do asynchronous read-ahead from an input io.Reader
+// and make the data available as an io.Reader.
+//
+// This should be fully transparent, except that once an error
+// has been returned from the Reader, it will not recover.
+//
+// The readahead object also fulfills the io.WriterTo interface, which
+// is likely to speed up copies.
+//
+// Package home: https://github.com/klauspost/readahead
+//
+package readahead
+
+import (
+	"errors"
+	"fmt"
+	"io"
+)
+
+type seekable struct {
+	*reader
+}
+
+type ReadSeekCloser interface {
+	io.ReadCloser
+	io.Seeker
+}
+
+type reader struct {
+	in      io.Reader     // Input reader
+	closer  io.Closer     // Optional closer
+	ready   chan *buffer  // Buffers ready to be handed to the reader
+	reuse   chan *buffer  // Buffers to reuse for input reading
+	exit    chan struct{} // Closes when finished
+	buffers int           // Number of buffers
+	size    int           // Size of each buffer
+	err     error         // If an error has occurred it is here
+	cur     *buffer       // Current buffer being served
+	exited  chan struct{} // Channel is closed been the async reader shuts down
+}
+
+// NewReaderSize returns a reader with a custom number of buffers and size.
+// buffers is the number of queued buffers and size is the size of each
+// buffer in bytes.
+func NewReaderSize(rd io.Reader, buffers, size int) (res io.ReadCloser, err error) {
+	if size <= 0 {
+		return nil, fmt.Errorf("buffer size too small")
+	}
+	if buffers <= 0 {
+		return nil, fmt.Errorf("number of buffers too small")
+	}
+	if rd == nil {
+		return nil, fmt.Errorf("nil input reader supplied")
+	}
+	a := &reader{}
+	if _, ok := rd.(io.Seeker); ok {
+		res = &seekable{a}
+	} else {
+		res = a
+	}
+	a.init(rd, buffers, size)
+	return
+}
+
+// initialize the reader
+func (a *reader) init(rd io.Reader, buffers, size int) {
+	a.in = rd
+	a.ready = make(chan *buffer, buffers)
+	a.reuse = make(chan *buffer, buffers)
+	a.exit = make(chan struct{}, 0)
+	a.exited = make(chan struct{}, 0)
+	a.buffers = buffers
+	a.size = size
+	a.cur = nil
+	a.err = nil
+
+	// Create buffers
+	for i := 0; i < buffers; i++ {
+		a.reuse <- newBuffer(size)
+	}
+
+	// Start async reader
+	go func() {
+		// Ensure that when we exit this is signalled.
+		defer close(a.exited)
+		defer close(a.ready)
+		for {
+			select {
+			case b := <-a.reuse:
+				err := b.read(a.in)
+				a.ready <- b
+				if err != nil {
+					return
+				}
+			case <-a.exit:
+				return
+			}
+		}
+	}()
+}
+
+// fill will check if the current buffer is empty and fill it if it is.
+// If an error was returned at the end of the current buffer it is returned.
+func (a *reader) fill() (err error) {
+	if a.cur.isEmpty() {
+		if a.cur != nil {
+			a.reuse <- a.cur
+			a.cur = nil
+		}
+		b, ok := <-a.ready
+		if !ok {
+			if a.err == nil {
+				a.err = errors.New("readahead: read after Close")
+			}
+			return a.err
+		}
+		a.cur = b
+	}
+	return nil
+}
+
+// Read will return the next available data.
+func (a *reader) Read(p []byte) (n int, err error) {
+	if a.err != nil {
+		return 0, a.err
+	}
+	// Swap buffer and maybe return error
+	err = a.fill()
+	if err != nil {
+		return 0, err
+	}
+
+	// Copy what we can
+	n = copy(p, a.cur.buffer())
+	a.cur.inc(n)
+
+	// If at end of buffer, return any error, if present
+	if a.cur.isEmpty() {
+		a.err = a.cur.err
+		return n, a.err
+	}
+	return n, nil
+}
+
+func (a *seekable) Seek(offset int64, whence int) (res int64, err error) {
+	//Not checking the result as seekable receiver guarantees it to be assertable
+	seeker, _ := a.in.(io.Seeker)
+	//Make sure the async routine is closed
+	select {
+	case <-a.exited:
+	case a.exit <- struct{}{}:
+		<-a.exited
+	}
+	if whence == io.SeekCurrent {
+		//If need to seek based on current position, take into consideration the bytes we read but the consumer
+		//doesn't know about
+		err = nil
+		for a.cur != nil {
+			if err = a.fill(); err == nil && a.cur != nil {
+				offset -= int64(len(a.cur.buffer()))
+				a.cur.offset = len(a.cur.buf)
+			}
+		}
+	}
+	//Seek the actual Seeker
+	if res, err = seeker.Seek(offset, whence); err == nil {
+		//If the seek was successful, reinitalize ourselves (with the new position).
+		a.init(a.in, a.buffers, a.size)
+	}
+	return
+}
+
+// WriteTo writes data to w until there's no more data to write or when an error occurs.
+// The return value n is the number of bytes written.
+// Any error encountered during the write is also returned.
+func (a *reader) WriteTo(w io.Writer) (n int64, err error) {
+	if a.err != nil {
+		return 0, a.err
+	}
+	n = 0
+	for {
+		err = a.fill()
+		if err != nil {
+			return n, err
+		}
+		n2, err := w.Write(a.cur.buffer())
+		a.cur.inc(n2)
+		n += int64(n2)
+		if err != nil {
+			return n, err
+		}
+		if a.cur.err != nil {
+			// io.Writer should return nil if we are at EOF.
+			if a.cur.err == io.EOF {
+				a.err = a.cur.err
+				return n, nil
+			}
+			a.err = a.cur.err
+			return n, a.cur.err
+		}
+	}
+}
+
+// Close will ensure that the underlying async reader is shut down.
+// It will also close the input supplied on newAsyncReader.
+func (a *reader) Close() (err error) {
+	select {
+	case <-a.exited:
+	case a.exit <- struct{}{}:
+		<-a.exited
+	}
+	if a.closer != nil {
+		// Only call once
+		c := a.closer
+		a.closer = nil
+		return c.Close()
+	}
+	a.err = errors.New("readahead: read after Close")
+	return nil
+}
+
+// Internal buffer representing a single read.
+// If an error is present, it must be returned
+// once all buffer content has been served.
+type buffer struct {
+	buf    []byte
+	err    error
+	offset int
+	size   int
+}
+
+func newBuffer(size int) *buffer {
+	return &buffer{buf: make([]byte, size), err: nil, size: size}
+}
+
+// isEmpty returns true is offset is at end of
+// buffer, or if the buffer is nil
+func (b *buffer) isEmpty() bool {
+	if b == nil {
+		return true
+	}
+	if len(b.buf)-b.offset <= 0 {
+		return true
+	}
+	return false
+}
+
+// read into start of the buffer from the supplied reader,
+// resets the offset and updates the size of the buffer.
+// Any error encountered during the read is returned.
+func (b *buffer) read(rd io.Reader) (err error) {
+	defer func() {
+		if r := recover(); r != nil {
+			err = fmt.Errorf("panic reading: %v", r)
+			b.err = err
+		}
+	}()
+	var n int
+	n, b.err = rd.Read(b.buf[0:b.size])
+	b.buf = b.buf[0:n]
+	b.offset = 0
+	return b.err
+}
+
+// Return the buffer at current offset
+func (b *buffer) buffer() []byte {
+	return b.buf[b.offset:]
+}
+
+// inc will increment the read offset
+func (b *buffer) inc(n int) {
+	b.offset += n
+}
diff --git a/s2/cmd/s2c/main.go b/s2/cmd/s2c/main.go
index 1481e34f0f..339e54f9b0 100644
--- a/s2/cmd/s2c/main.go
+++ b/s2/cmd/s2c/main.go
@@ -17,6 +17,7 @@ import (
 	"unicode"
 
 	"github.com/klauspost/compress/s2"
+	"github.com/klauspost/compress/s2/cmd/internal/readahead"
 )
 
 var (
@@ -88,6 +89,9 @@ Options:`)
 		func() {
 			var closeOnce sync.Once
 			dstFilename := fmt.Sprintf("%s%s", filename, ".s2")
+			if *bench > 0 {
+				dstFilename = "(discarded)"
+			}
 			if !*quiet {
 				fmt.Println("Compressing", filename, "->", dstFilename)
 			}
@@ -95,7 +99,9 @@ Options:`)
 			file, err := os.Open(filename)
 			exitErr(err)
 			defer closeOnce.Do(func() { file.Close() })
-			src := bufio.NewReaderSize(file, int(sz)*2)
+			src, err := readahead.NewReaderSize(file, *cpu, int(sz))
+			exitErr(err)
+			defer src.Close()
 			finfo, err := file.Stat()
 			exitErr(err)
 			var out io.Writer
diff --git a/s2/cmd/s2d/main.go b/s2/cmd/s2d/main.go
index c2bf8761f2..70004c671a 100644
--- a/s2/cmd/s2d/main.go
+++ b/s2/cmd/s2d/main.go
@@ -14,6 +14,7 @@ import (
 	"time"
 
 	"github.com/klauspost/compress/s2"
+	"github.com/klauspost/compress/s2/cmd/internal/readahead"
 )
 
 var (
@@ -78,6 +79,9 @@ Options:`)
 			fmt.Println("Skipping", filename)
 			continue
 		}
+		if *bench > 0 {
+			dstFilename = "(discarded)"
+		}
 
 		func() {
 			var closeOnce sync.Once
@@ -89,7 +93,9 @@ Options:`)
 			exitErr(err)
 			defer closeOnce.Do(func() { file.Close() })
 			rc := rCounter{in: file}
-			src := bufio.NewReaderSize(&rc, 4<<20)
+			src, err := readahead.NewReaderSize(&rc, 2, 4<<20)
+			exitErr(err)
+			defer src.Close()
 			finfo, err := file.Stat()
 			exitErr(err)
 			mode := finfo.Mode() // use the same mode for the output file