by Sathish Palaniappan, Pramod Nagaraja | Published September 2, 2008
Many Web applications serve a significant amount of static content, which amounts to reading data off of a disk and writing the exact same data back to the response socket. This activity might appear to require relatively little CPU activity, but it’s somewhat inefficient: the kernel reads the data off of disk and pushes it across the kernel-user boundary to the application, and then the application pushes it back across the kernel-user boundary to be written out to the socket. In effect, the application serves as an inefficient intermediary that gets the data from the disk file to the socket.
Each time data traverses the user-kernel boundary, it must be copied, which consumes CPU cycles and memory bandwidth. Fortunately, you can eliminate these copies through a technique called — appropriately enough —zero copy. Applications that use zero copy request that the kernel copy the data directly from the disk file to the socket, without going through the application. Zero copy greatly improves application performance and reduces the number of context switches between kernel and user mode.
The Java class libraries support zero copy on Linux and UNIX systems through the transferTo() method in java.nio.channels.FileChannel. You can use the transferTo() method to transfer bytes directly from the channel on which it is invoked to another writable byte channel, without requiring data to flow through the application. This article first demonstrates the overhead incurred by simple file transfer done through traditional copy semantics, then shows how the zero-copy technique using transferTo() achieves better performance.
Consider the scenario of reading from a file and transferring the data to another program over the network. (This scenario describes the behavior of many server applications, including Web applications serving static content, FTP servers, mail servers, and so on.) The core of the operation is in the two calls in Listing 1 (download the complete sample code):
File.read(fileDesc, buf, len);
Socket.send(socket, buf, len);
Although Listing 1 is conceptually simple, internally, the copy operation requires four context switches between user mode and kernel mode, and the data is copied four times before the operation is complete. Figure 1 shows how data is moved internally from the file to the socket:
Figure 2 shows the context switching:
The steps involved are:
Use of the intermediate kernel buffer (rather than a direct transfer of the data into the user buffer) might seem inefficient. But intermediate kernel buffers were introduced into the process to improve performance. Using the intermediate buffer on the read side allows the kernel buffer to act as a “readahead cache” when the application hasn’t asked for as much data as the kernel buffer holds. This significantly improves performance when the requested data amount is less than the kernel buffer size. The intermediate buffer on the write side allows the write to complete asynchronously.
Unfortunately, this approach itself can become a performance bottleneck if the size of the data requested is considerably larger than the kernel buffer size. The data gets copied multiple times among the disk, kernel buffer, and user buffer before it is finally delivered to the application.
Zero copy improves performance by eliminating these redundant data copies.
If you re-examine the traditional scenario, you’ll notice that the second and third data copies are not actually required. The application does nothing other than cache the data and transfer it back to the socket buffer. Instead, the data could be transferred directly from the read buffer to the socket buffer. The transferTo() method lets you do exactly this. Listing 2 shows the method signature of transferTo():
public void transferTo(long position, long count, WritableByteChannel target);
The transferTo() method transfers data from the file channel to the given writable byte channel. Internally, it depends on the underlying operating system’s support for zero copy; in UNIX and various flavors of Linux, this call is routed to the sendfile() system call, shown in Listing 3, which transfers data from one file descriptor to another:
ssize_t sendfile(int out_fd, int in_fd, off_t *offset, size_t count);
The action of the file.read() and socket.send() calls in Listing 1 can be replaced by a single transferTo() call, as shown in Listing 4:
transferTo(position, count, writableChannel);
Figure 3 shows the data path when the transferTo() method is used:
Figure 4 shows the context switches when the transferTo() method is used:
The steps taken when you use transferTo() as in Listing 4 are:
This is an improvement: we’ve reduced the number of context switches from four to two and reduced the number of data copies from four to three (only one of which involves the CPU). But this does not yet get us to our goal of zero copy. We can further reduce the data duplication done by the kernel if the underlying network interface card supports gather operations. In Linux kernels 2.4 and later, the socket buffer descriptor was modified to accommodate this requirement. This approach not only reduces multiple context switches but also eliminates the duplicated data copies that require CPU involvement. The user-side usage still remains the same, but the intrinsics have changed:
Figure 5 shows the data copies using transferTo() with the gather operation:
Now let’s put zero copy into practice, using the same example of transferring a file between a client and a server (see Download for the sample code). TraditionalClient.java and TraditionalServer.java are based on the traditional copy semantics, using File.read() and Socket.send(). TraditionalServer.java is a server program that listens on a particular port for the client to connect, and then reads 4K bytes of data at a time from the socket. TraditionalClient.java connects to the server, reads (using File.read()) 4K bytes of data from a file, and sends (using socket.send()) the contents to the server via the socket.
Similarly, TransferToServer.java and TransferToClient.java perform the same function, but instead use the transferTo() method (and in turn the sendfile() system call) to transfer the file from server to client.
We executed the sample programs on a Linux system running the 2.6 kernel and measured the run time in milliseconds for both the traditional approach and the transferTo() approach for various sizes. Table 1 shows the results:
As you can see, the transferTo() API brings down the time approximately 65 percent compared to the traditional approach. This has the potential to increase performance significantly for applications that do a great deal of copying of data from one I/O channel to another, such as Web servers.
We have demonstrated the performance advantages of using transferTo() compared to reading from one channel and writing the same data to another. Intermediate buffer copies — even those hidden in the kernel — can have a measurable cost. In applications that do a great deal of copying of data between channels, the zero-copy technique can offer a significant performance improvement.
April 11, 2019
March 19, 2019
June 20, 2019
Back to top