How does the new I/O (NIO) offer better scalability and better performance?

Java has long been not suited for developing programs that perform a lot of I/O operations. Furthermore, commonly needed tasks such as file locking, non-blocking and asynchronous I/O operations and ability to map file to memory were not available. Non-blocking I/O operations were achieved through work around such as multithreading or using JNI. The New I/O API (aka NIO) in J2SE 1.4 has changed this situation.

A server’s ability to handle several client requests effectively depends on how it uses I/O streams. When a server has to handle hundreds of clients simultaneously, it must be able to use I/O services concurrently. One way to cater for this scenario in Java is to use threads but having almost one-to-one ratio of threads (100 clients will have 100 threads) is prone to enormous thread overhead and can result in performance and scalability problems due to consumption of memory stacks and CPU context switching (i.e. switching between threads as opposed to doing real computation.). To overcome this problem, a new set of non-blocking I/O classes have been introduced to the Java platform in java.nio package. The non-blocking I/O mechanism is built around Selectors and Channels. Channels, Buffers and Selectors are the core of the NIO.

A Channel class represents a bi-directional communication channel (similar to InputStream and OutputStream) between datasources such as a socket, a file, or an application component, which is capable of performing one or more I/O operations such as reading or writing. Channels can be non-blocking, which means, no I/O operation will wait for data to be read or written to the network. The good thing about NIO channels is that they can be asynchronously interrupted and closed. So if a thread is blocked in an I/O operation on a channel, another thread can interrupt that blocked thread.

A Selector class enables multiplexing (combining multiple streams into a single stream) and demultiplexing (separating a single stream into multiple streams) I/O events and makes it possible for a single thread to efficiently manage many I/O channels. A Selector monitors selectable channels, which are registered with it for I/O events like connect, accept, read and write. The keys (i.e. Key1, Key2 etc represented by the SelectionKey class) encapsulate the relationship between a specific selectable channel and a specific selector.

Buffers hold data. Channels can fill and drain Buffers. Buffers replace the need for you to do your own buffer management using byte arrays. There are different types of Buffers like ByteBuffer, CharBuffer, DoubleBuffer, etc.

Design pattern: NIO uses a reactor design pattern, which demultiplexes events (separating single stream into multiple streams) and dispatches them to registered object handlers. The reactor pattern is similar to an observer pattern (aka publisher and subscriber design pattern), but an observer pattern handles only a single source of events (i.e. a single publisher with multiple subscribers) where a reactor pattern handles multiple event sources (i.e. multiple publishers with multiple subscribers). The intent of an observer pattern is to define a one-to-many dependency so that when one object (i.e. the publisher) changes its state, all its dependents (i.e. all its subscribers) are notified and updated correspondingly.

Another sought after functionality of NIO is its ability to map a file to memory. There is a specialized form of a Buffer known as “MappedByteBuffer”, which represents a buffer of bytes mapped to a file. To map a file to “MappedByteBuffer”, you must first get a channel for a file. Once you get a channel then you map it to a buffer and subsequently you can access it like any other “ByteBuffer”. Once you map an input file to a “CharBuffer”, you can do pattern matching on the file contents. This is similar to running “grep” on a UNIX file system.

Another feature of NIO is its ability to lock and unlock files. Locks can be exclusive or shared and can be held on a contiguous portion of a file. But file locks are subject to the control of the underlying operating system.

Tagged . Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.