Homework 1

Due Date

April 11th 2005: 5:00PM

Description

You will implement a client-server system which operates a bank. There will be a single server which will maintain a single account. Intially the account balance is 0. There can be many clients who can request service simultaneously from this server. The services are of three types:

Deposit : The client might make a deposit into the account and the account will be credited that much amount. The deposit must be a positive floating point number. The server will acknowledge successful deposit by returning the current balance to the client.

Withdraw : The client can make a withdrawal of some positive floating amount. The account will be debited that much amount. If the account does not have sufficient balance, then an error should be returned to the client, otherwise the current balance is returned.

Balance : The client can ask for the balance and the server should reply with the current account balance.

Requirements

All code must be written in C. Network communication will use reliable TCP sockets.

Server

The server will be called server and invoked as follows

 $ server -p port -d delay

The arguments can be specified in any order. To parse arguments, use getopt. See man 3 getopt for more details. By default if port is not specified then server will listen on port 8000. The delay is an integer which specifies the time in milliseconds required to complete the deposit/withdraw/balance operations. You will be holding the mutex lock during this delay. By default the delay is 100msec. Use usleep to implement delay (see man 3 usleep) for details.

Client

The client will be called client and invoked as follows:

$ client -h servername -p port -d deposit -w withdraw -b

The servername specifies which computer is running the server and is compulsory. If port is not specified, use port 8000. Only one of -d, -w and -b options can be specified. After the client is done, it should print the balance returned by the server on stdout or -1 if an error occurred.

Other Issues

All code must be written in C on the csil machines. We shall compile and run code there. If you do not have an account there, then please go to the CS office and get an account.
For a tutorial on sockets see Socket client server tutorial from RPI. For more useful info, also see man 7 socket, man 7 ip
Sun Microsystems has an excellent multithreaded programming guide , which discusses multithreaded programming using pthreads and Solaris native threads. The pthread interface for linux is described in the man pages as well.
Floating point and integer representations differ from architecture to architecture. So you can't just transmit the floating point number through the network without converting it into a machine independent format. The simplest and most inefficient way is to write it out into a char string and send the char buffer. A better and more compact way is to use Sun's External Data Representation standard. See man xdr for details. The choice is yours.
For every request by a client, the server must spawn a new thread to handle it. To make sure that the account balance is always consistent, the balance must be protected with a mutex lock. Remember that the balance operation requires a read lock on the account balance, while deposit and withdraw operations require a write lock. Since we want the implementation to have high levels of concurrency, you will have to use reader-writer locks. pthreads do have a reader writer lock interface, but as of now, linux doesn't seem to have an implementation of them. Most OS books tell you how to implement reader-writer locks using simple mutex locks.
If you are lazy, you can use the pthread read-write locks. To do this either add the line
#define _GNU_SOURCE
at the top of your file before including any headers or add
-D_GNU_SOURCE
to the C compiler options.
```
gcc -D_GNU_SOURCE my.c -lpthread
```
The pthread read write locks are documented in the Sun multithreaded programming guide mentioned above.