Networking and Socket Programming is one of the important area of Java programming language, especially for those programmers, who are working in client server based applications. Knowledge of important protocols e.g. TCP and UDP in detail is very important, especially if you are in business of writing high frequency trading application, which communicate via FIX Protocol or native exchange protocol. In this article, we will some of the frequently asked questions on networking and socket programming, mostly based around TCP IP protocol. This article is kinda light on NIO though, as it doesn't include questions from multiplexing, selectors, ByteBuffer and FileChannel but it does include classical questions like difference between IO and NIO. Main focus of this post is to make Java developer familiar with low level parts e.g. how TCP and UDP protocol works, socket options and writing multi-threaded servers in Java. Questions discussed here is not really tied up with Java programming language, and can be used in any programming language, which allows programmers to write client-server applications. By the way, If you are going for interview on Investment banks for core Java developer role, you better prepare well on Java NIO, Socket Programming, TCP, UDP and Networking along with other popular topics e.g. multi-threading, Collections API and Garbage Collection tuning. You can also contribute any question, which is asked to you or related to socket programming and networking and can be useful for Java interviews.
1) Difference between TCP and UDP protocol?
There are many differences between TCP (Transmission control Protocol) and UDP (User Datagram Protocol), but main is TCP is connection oriented, while UDP is connection less. This means TCP provides guaranteed delivery of messages in the order they are sent, while UDP doesn't provide any delivery guarantee. Because of this guarantee, TCP is slower than UDP, as it needs to perform more work. TCP is best suited for message, which you can't afford to loss, e.g. order and trade messages in electronic trading, wire transfer in banking and finance etc. UDP is more suited for media transmission, where loss of one packet, known as datagrams is affordable and doesn't affect quality of service. This answer is enough for most of the interviews, but you need to be more detailed when you are interviewing as Java developer for high frequency trading desk. Some of the points which many candidate forget to mention is about order and data boundary. In TCP, messages are guaranteed to be delivered in the same order as they are sent but data boundary is not preserved, which means multiple messages can be combined and sent together, or receiver may receive one part of the message in one packet and other part of the message in next packet. Though application will receive full message and in the same order. TCP protocol will do assembling of message for you. On the other hand, UDP sends full message in a datagram packet, if clients receives the packet it is guaranteed that it will get the full message, but there is no guarantee that packet will come in same order they are sent. In short, you must mention following differences between TCP and UDP protocol while answering during interview :
for more detailed answer, see my post 9 differences between TCP and UDP protocol.
2) How does TCP handshake works?
Three messages are exchanged as part of TCP head-shake e.g. Initiator sends SYN, upon receiving this Listener sends SYN-ACK, and finally initiator replied with ACK, at this point TCP connection is moved to ESTABLISHED state. This process is easily understandable by looking at following diagram.
3) How do you implement reliable transmission in UDP protocol?
This is usually follow-up of previous interview question. Though UDP doesn't provide delivery guarantee at protocol level, you can introduce your own logic to maintain reliable messaging e.g. by introducing sequence numbers and retransmission. If receiver find that it has missed a sequence number, it can ask for replay of that message from Server. TRDP protocol, which is used Tibco Rendezvous (a popular high speed messaging middle-ware) uses UDP for faster messaging and provides reliability guarantee by using sequence number and retransmission.
4) What is Network Byte Order? How does two host communicate if they have different byte-ordering?
There are two ways to store two bytes in memory, little endian (least significant byte at the starting address) and big endian (most significant byte at the starting address). They are collectively known as host byte order. For example, an Intel processor stores the 32-bit integer as four consecutive bytes in memory in the order 1-2-3-4, where 1 is the most significant byte. IBM PowerPC processors would store the integer in the byte order 4-3-2-1. Networking protocols such as TCP are based on a specific network byte order, which uses big-endian byte ordering. If two machines are communicating with each other and they have different byte ordering, they are converted to network byte order before sending or after receiving. Therefore, a little endian micro-controller sending to a UDP/IP network must swap the order in which bytes appear within multi byte values before the values are sent onto the network, and must swap the order in which bytes appear in multi byte values received from the network before the values are used. In short, you can also say network byte order is standard of storing byte during transmission, and it uses big endian byte ordering mechanism.
5) What is Nagle's algorithm?
If interviewer is testing your knowledge of TCP/IP protocol than it's very rare for him not to ask this question. Nagle's algorithm is way of improving performance of TCP/IP protocol and networks by reducing number of TCP packets that needs to be sent over network. It works by buffering small packets until buffer reaches Maximum Segment Size. Since small packets, which contains only 1 or 2 bytes of data, has more overhead in terms of TCP header, which is of 40 bytes. These small packets can also leads to congestion in slow network. Nagle's algorithm tries to improve efficiency of TCP protocol by buffering them, to send a larger packet. Also Nagle's algorithm has negative effect on non small writes, so if you are writing large data on packets than it's better to disable Nagle's algorithm. In general, Nagle's algorithm is a defence against careless application, which sends lots of small packets to network, but it will not benefit or have a negative effect on well written application, which properly takes care of buffering.
6) What is TCP_NODELAY?
TCP_NODELAY is an option to disable Nagle's algorithm, provided by various TCP implementations. Since Nagle's algorithm performs badly with TCP delayed acknowledgement algorithm, it's better to disable Nagle's when you are doing write-write-read operation. Where a read after two successive write on socket may get delayed up-to 500 millisecond, until the second write has reached the destination. If latency is more concern over bandwidth usage e.g. in a network based multi-player game, user wants to see action from other player immediately, it's better to bypass Nagle's delay by using TCP_NODELAY flag.
7) What is multicasting or multicast transmission? Which Protocol is generally used for multicast? TCP or UDP?
Multi-casting or multicast transmission is one to many distribution, where message is delivered to a group of subscribers simultaneously in a single transmission from publisher. Copies of messages are automatically created in other network elements e.g. Routers, but only when the topology of network requires it. Tibco Rendezvous supports multicast transmission. Multi-casting can only be implemented using UDP, because it sends full data as datagram package, which can be replicated and delivered to other subscribers. Since TCP is a point-to-point protocol, it can not deliver messages to multiple subscriber, until it has link between each of them. Though, UDP is not reliable, and messages may be lost or delivered out of order. Reliable multicast protocols such as Pragmatic General Multicast (PGM) have been developed to add loss detection and retransmission on top of IP multicast. IP multicast is widely deployed in enterprises, commercial stock exchanges, and multimedia content delivery networks. A common enterprise use of IP multicast is for IPTV applications
8) What is difference between Topic and Queue in JMS?
Main difference between Topic and Queue in Java Messaging Service comes when we have multiple consumers to consumer messages. If we set-up multiple listener thread to consume messages from Queue, each messages will be dispatched to only one thread and not all thread. On the other hand in case of Topic each subscriber gets it's own copy of message.
9) What is difference between IO and NIO?
Main difference between NIO and IO is that NIO provides asynchronous, non blocking IO, which is critical to write faster and scalable networking systems. While most of utility from IO classes are blocking and slow. NIO take advantage of asynchronous system calls in UNIX systems such as select() system call for network sockets. Using select(), an application can monitor several resources at the same time and can also poll for network activity without blocking. The select() system call identifies if data is pending or not, then read() or write() may be used knowing that they will complete immediately.
10) How do you write multi-threaded server in Java?
A multi-threaded server is the one which can server multiple clients without blocking. Java provides excellent support to developer such server. Prior to Java 1.4, you can write multi-threaded server using traditional socket IO and threads. This had severe limitation on scalability, because it creates new thread for each connection and you can only create a fixed number of threads, depending upon machine's and platform's capability. Though this design can be improved by using thread pools and worker threads, it still a resource intensive design. After JDK 1.4 and NIO's introduction, writing scalable and multi-threaded server become bit easier. You can easily create it in single thread by using Selector, which takes advantage of asynchronous and non-blocking IO model of Java NIO.
11) What is ephemeral port?
In TCP/IP connection usually contains four things, Server IP, Server port, Client IP and Client Port. Out of these four, 3 are well known in most of the time, what is not known is client port, this is where ephemeral ports comes into picture. ephemeral ports are dynamic port assigned by your machine's IP stack, from a specified range, known as ephemeral port range, when a client connection explicitly doesn't specify a port number. These are short lived, temporary port, which can be reused once connection is closed, but most of IP software, doesn't reuse ephemeral port, until whole range is exhausted. Similar to TCP, UDP protocol also uses ephemeral port, while sending datagram . In Linux ephemeral port range is from 32768 to 61000, while in windows default ephemeral port range is 1025 to 5000. Similarly different operating system has different ephemeral port ranges
12) What is sliding window protocol?
Sliding window protocol is a technique for controlling transmitted data packets between two network computers where reliable and sequential delivery of data packets is required, such as provided by Transmission Control Protocol (TCP). In the sliding window technique, each packet includes a unique consecutive sequence number, which is used by the receiving computer to place data in the correct order. The objective of the sliding window technique is to use the sequence numbers to avoid duplicate data and to request missing data
13) When do you get "too many files open" error?
Just like File connection, Socket Connection also needs file descriptors, Since every machine has limited number of file descriptors, it's possible that they may ran out of file descriptors. When it happen, you will see "too many files open" error. You can check how many file descriptor per process is allowed on UNIX based system by executing ulimit -n command or simply count entries on /proc//fd/
14) What is TIME_WAIT state in TCP protocol? When does a socket connection goes to TIME_WAIT state?
When one end of TCP Connection closes it by making system call, it goes into TIME_WAIT state. Since TCP packets can arrive in wrong order, the port must not be closed immediately to allow late packets to arrive. That's why that end of TCP connection goes into TIME_WAIT state. For example, if client closes a socket connection than it will go to TIME_WAIT state, similarly if server closes connection than you will see TIME_WAIT there. You can check status of your TCP and UDP sockets by using these networking commands in UNIX.
15) What will happen if you have too many socket connections in TIME_WAIT state on Server?
When a socket connection or port goes into TIME_WAIT state, it doesn't release file descriptor associated with it. File descriptor is only released when TIME_WAIT state is gone i.e. after some specified configured time. If too many connections are in TIME_WAIT state than your Server may ran out of file descriptors and start throwing "too many files open" error, and stop accepting new connections.
That's all about in this list of networking and socket programming interview questions and answers. Though I have originally intended this list for Java programmers it is equally useful for any programmer. In fact, this is bare minimum knowledge of sockets and protocols every programmer should have. I have found that C and C++ programmers are better answering these questions than an average Java programmer. One reason of this may be because Java programmers has got so many useful library e.g. Apache MINA, which does all the low level work for them. Anyway, knowledge of fundamentals is very important and everything else is just an excuse, but at same point I also recommend using tried and tested libraries like Apache MINA for production code.
Java Networking and Socket Programming Questions Answers
Here is my list of 15 interview questions related to networking basics, internet protocol and socket programming in Java. Though it doesn't contain basic questions form API e.g. Server, ServerSocket, but it focus on high level concept of writing scalable server in Java using NIO selectors and how to implement that using threads, there limitations and issues etc. I will probably add few more questions based on some best practices while writing socket based application in Java. If you know a good question on this topic, feel free to suggest.1) Difference between TCP and UDP protocol?
There are many differences between TCP (Transmission control Protocol) and UDP (User Datagram Protocol), but main is TCP is connection oriented, while UDP is connection less. This means TCP provides guaranteed delivery of messages in the order they are sent, while UDP doesn't provide any delivery guarantee. Because of this guarantee, TCP is slower than UDP, as it needs to perform more work. TCP is best suited for message, which you can't afford to loss, e.g. order and trade messages in electronic trading, wire transfer in banking and finance etc. UDP is more suited for media transmission, where loss of one packet, known as datagrams is affordable and doesn't affect quality of service. This answer is enough for most of the interviews, but you need to be more detailed when you are interviewing as Java developer for high frequency trading desk. Some of the points which many candidate forget to mention is about order and data boundary. In TCP, messages are guaranteed to be delivered in the same order as they are sent but data boundary is not preserved, which means multiple messages can be combined and sent together, or receiver may receive one part of the message in one packet and other part of the message in next packet. Though application will receive full message and in the same order. TCP protocol will do assembling of message for you. On the other hand, UDP sends full message in a datagram packet, if clients receives the packet it is guaranteed that it will get the full message, but there is no guarantee that packet will come in same order they are sent. In short, you must mention following differences between TCP and UDP protocol while answering during interview :
- TCP is guaranteed delivery, UDP is not guaranteed.
- TCP guarantees order of messages, UDP doesn't.
- Data boundary is not preserved in TCP, but UDP preserves it.
- TCP is slower compared to UDP.
for more detailed answer, see my post 9 differences between TCP and UDP protocol.
2) How does TCP handshake works?
Three messages are exchanged as part of TCP head-shake e.g. Initiator sends SYN, upon receiving this Listener sends SYN-ACK, and finally initiator replied with ACK, at this point TCP connection is moved to ESTABLISHED state. This process is easily understandable by looking at following diagram.
3) How do you implement reliable transmission in UDP protocol?
This is usually follow-up of previous interview question. Though UDP doesn't provide delivery guarantee at protocol level, you can introduce your own logic to maintain reliable messaging e.g. by introducing sequence numbers and retransmission. If receiver find that it has missed a sequence number, it can ask for replay of that message from Server. TRDP protocol, which is used Tibco Rendezvous (a popular high speed messaging middle-ware) uses UDP for faster messaging and provides reliability guarantee by using sequence number and retransmission.
4) What is Network Byte Order? How does two host communicate if they have different byte-ordering?
There are two ways to store two bytes in memory, little endian (least significant byte at the starting address) and big endian (most significant byte at the starting address). They are collectively known as host byte order. For example, an Intel processor stores the 32-bit integer as four consecutive bytes in memory in the order 1-2-3-4, where 1 is the most significant byte. IBM PowerPC processors would store the integer in the byte order 4-3-2-1. Networking protocols such as TCP are based on a specific network byte order, which uses big-endian byte ordering. If two machines are communicating with each other and they have different byte ordering, they are converted to network byte order before sending or after receiving. Therefore, a little endian micro-controller sending to a UDP/IP network must swap the order in which bytes appear within multi byte values before the values are sent onto the network, and must swap the order in which bytes appear in multi byte values received from the network before the values are used. In short, you can also say network byte order is standard of storing byte during transmission, and it uses big endian byte ordering mechanism.
5) What is Nagle's algorithm?
If interviewer is testing your knowledge of TCP/IP protocol than it's very rare for him not to ask this question. Nagle's algorithm is way of improving performance of TCP/IP protocol and networks by reducing number of TCP packets that needs to be sent over network. It works by buffering small packets until buffer reaches Maximum Segment Size. Since small packets, which contains only 1 or 2 bytes of data, has more overhead in terms of TCP header, which is of 40 bytes. These small packets can also leads to congestion in slow network. Nagle's algorithm tries to improve efficiency of TCP protocol by buffering them, to send a larger packet. Also Nagle's algorithm has negative effect on non small writes, so if you are writing large data on packets than it's better to disable Nagle's algorithm. In general, Nagle's algorithm is a defence against careless application, which sends lots of small packets to network, but it will not benefit or have a negative effect on well written application, which properly takes care of buffering.
6) What is TCP_NODELAY?
TCP_NODELAY is an option to disable Nagle's algorithm, provided by various TCP implementations. Since Nagle's algorithm performs badly with TCP delayed acknowledgement algorithm, it's better to disable Nagle's when you are doing write-write-read operation. Where a read after two successive write on socket may get delayed up-to 500 millisecond, until the second write has reached the destination. If latency is more concern over bandwidth usage e.g. in a network based multi-player game, user wants to see action from other player immediately, it's better to bypass Nagle's delay by using TCP_NODELAY flag.
7) What is multicasting or multicast transmission? Which Protocol is generally used for multicast? TCP or UDP?
Multi-casting or multicast transmission is one to many distribution, where message is delivered to a group of subscribers simultaneously in a single transmission from publisher. Copies of messages are automatically created in other network elements e.g. Routers, but only when the topology of network requires it. Tibco Rendezvous supports multicast transmission. Multi-casting can only be implemented using UDP, because it sends full data as datagram package, which can be replicated and delivered to other subscribers. Since TCP is a point-to-point protocol, it can not deliver messages to multiple subscriber, until it has link between each of them. Though, UDP is not reliable, and messages may be lost or delivered out of order. Reliable multicast protocols such as Pragmatic General Multicast (PGM) have been developed to add loss detection and retransmission on top of IP multicast. IP multicast is widely deployed in enterprises, commercial stock exchanges, and multimedia content delivery networks. A common enterprise use of IP multicast is for IPTV applications
8) What is difference between Topic and Queue in JMS?
Main difference between Topic and Queue in Java Messaging Service comes when we have multiple consumers to consumer messages. If we set-up multiple listener thread to consume messages from Queue, each messages will be dispatched to only one thread and not all thread. On the other hand in case of Topic each subscriber gets it's own copy of message.
9) What is difference between IO and NIO?
Main difference between NIO and IO is that NIO provides asynchronous, non blocking IO, which is critical to write faster and scalable networking systems. While most of utility from IO classes are blocking and slow. NIO take advantage of asynchronous system calls in UNIX systems such as select() system call for network sockets. Using select(), an application can monitor several resources at the same time and can also poll for network activity without blocking. The select() system call identifies if data is pending or not, then read() or write() may be used knowing that they will complete immediately.
10) How do you write multi-threaded server in Java?
A multi-threaded server is the one which can server multiple clients without blocking. Java provides excellent support to developer such server. Prior to Java 1.4, you can write multi-threaded server using traditional socket IO and threads. This had severe limitation on scalability, because it creates new thread for each connection and you can only create a fixed number of threads, depending upon machine's and platform's capability. Though this design can be improved by using thread pools and worker threads, it still a resource intensive design. After JDK 1.4 and NIO's introduction, writing scalable and multi-threaded server become bit easier. You can easily create it in single thread by using Selector, which takes advantage of asynchronous and non-blocking IO model of Java NIO.
11) What is ephemeral port?
In TCP/IP connection usually contains four things, Server IP, Server port, Client IP and Client Port. Out of these four, 3 are well known in most of the time, what is not known is client port, this is where ephemeral ports comes into picture. ephemeral ports are dynamic port assigned by your machine's IP stack, from a specified range, known as ephemeral port range, when a client connection explicitly doesn't specify a port number. These are short lived, temporary port, which can be reused once connection is closed, but most of IP software, doesn't reuse ephemeral port, until whole range is exhausted. Similar to TCP, UDP protocol also uses ephemeral port, while sending datagram . In Linux ephemeral port range is from 32768 to 61000, while in windows default ephemeral port range is 1025 to 5000. Similarly different operating system has different ephemeral port ranges
12) What is sliding window protocol?
Sliding window protocol is a technique for controlling transmitted data packets between two network computers where reliable and sequential delivery of data packets is required, such as provided by Transmission Control Protocol (TCP). In the sliding window technique, each packet includes a unique consecutive sequence number, which is used by the receiving computer to place data in the correct order. The objective of the sliding window technique is to use the sequence numbers to avoid duplicate data and to request missing data
13) When do you get "too many files open" error?
Just like File connection, Socket Connection also needs file descriptors, Since every machine has limited number of file descriptors, it's possible that they may ran out of file descriptors. When it happen, you will see "too many files open" error. You can check how many file descriptor per process is allowed on UNIX based system by executing ulimit -n command or simply count entries on /proc/
14) What is TIME_WAIT state in TCP protocol? When does a socket connection goes to TIME_WAIT state?
When one end of TCP Connection closes it by making system call, it goes into TIME_WAIT state. Since TCP packets can arrive in wrong order, the port must not be closed immediately to allow late packets to arrive. That's why that end of TCP connection goes into TIME_WAIT state. For example, if client closes a socket connection than it will go to TIME_WAIT state, similarly if server closes connection than you will see TIME_WAIT there. You can check status of your TCP and UDP sockets by using these networking commands in UNIX.
15) What will happen if you have too many socket connections in TIME_WAIT state on Server?
When a socket connection or port goes into TIME_WAIT state, it doesn't release file descriptor associated with it. File descriptor is only released when TIME_WAIT state is gone i.e. after some specified configured time. If too many connections are in TIME_WAIT state than your Server may ran out of file descriptors and start throwing "too many files open" error, and stop accepting new connections.
That's all about in this list of networking and socket programming interview questions and answers. Though I have originally intended this list for Java programmers it is equally useful for any programmer. In fact, this is bare minimum knowledge of sockets and protocols every programmer should have. I have found that C and C++ programmers are better answering these questions than an average Java programmer. One reason of this may be because Java programmers has got so many useful library e.g. Apache MINA, which does all the low level work for them. Anyway, knowledge of fundamentals is very important and everything else is just an excuse, but at same point I also recommend using tried and tested libraries like Apache MINA for production code.
No comments:
Post a Comment