Prepared by Natalia Khuri and Seetharam Samptur

Chapter 9 Interprocess Communication

By Sape J. Mullender

Prepared by Natalia Khuri and Seetharam Samptur

Homework Questions and Answers

Q1.Easy: Give an example of communication where error correcting code works better than error detection and an example, where neither error detection nor error correction is needed.

A1.Easy: Error correcting code works better in the case of communication between a space shuttle and its land control station. Due to long distance, the error rates are high and so is the end-to-end latency. Data retransmission can be avoided by including error-correcting codes.

In interactive continuous media application, the maximum acceptable end-to-end latency is in the order of tens of milliseconds. Data retransmission is ruled out, as late data is useless in audio/video applications. Since, human eye and ear have a “built-in” error correction mechanism small gaps in data stream would not be noticeable thus eliminating the need for error correcting codes.

Q2.Easy: At-least-once protocols deliver messages exactly once in the absence of failure and may deliver messages more than once when failures occur. Such protocols work when requests are idempotent; that is, when carrying the request more than once has the same effect as carrying them out several times. Consider four possible file operations listed below. Determine which operations are idempotent and which are not.

• Reading a file
• Writing to a file
• Creating a file
• Deleting a file

A2.Easy: Reading and writing a data block once has the same effect as reading and writing it multiple times. Creating or deleting files are also idempotent, the result of the operation is the same even though the first time the operation will succeed and all subsequent tries will fail.

Q3.Easy: Certain transport protocols operate in the context of a session in order to maintain protocol state and to guarantee at-most-once message delivery. Sessions must have unique names, so that messages sent in one session are not received in another. Propose a session name generation protocol for creating unique and crush-resistant session names.

A3.Easy: A good way to choose session identifiers is to use timestamps as part of the identifier. Most modern processors have a battery-operated clock whose value survives crushes. Even, if such clock is missing, they can ask a network time service for the time when they come up.

Q4.Easy: In RPCs, a program running on a machine of one endianness may have to deal with structured data exported from a machine of a different endianness. Write a simple ‘C’ code snippet to determine a system’s endianness.

A4.Easy:

union u {

short s;

char c[sizeof(short)];

};

void sys_endianness()

{

union u u1;

u1.s = 0x0123;

if ( u1.c[0] == 0x23 )

printf("little-endian\n");

else

printf("big-endian\n");

}

int main(int argc, char *argv)

{

sys_endianness();

}

Q5.Difficult: Why is the process of multiplexing and demultiplexing data stream deemed important in network communications? How does ATM make this process efficient compared to Ethernet or other packet switched networks?

A.5.Difficult: There could be multiple applications communicating from a host and hence large amount of data will be transmitted and received simultaneously. Additionally, each application uses a separate network connection or virtual circuits for communicating with a remote server. Hence sending and receiving data involves lot of internal transfers or copies between the different protocol layers thus increasing end-to-end latency.

In ATM, each cell size is fixed at 53 bytes, of which the header is of 5 bytes and the rest user payload. The inclusion of the VCI in the header and due to the small cell size, the ATM switches and edge nodes can parse the information quickly. Also, since the VCI is associated with the application that initiated the connection, the ATM interface card can DMA the payload into the receiving process’ memory thus avoiding data copy.

In an Ethernet or token ring network, the MAC level packet headers do not contain any information to demultiplex the received information. Also, the packet size if around 1500 bytes when compared to the smaller 53-byte cell in an ATM network making it harder to parse in a short time.

Q6.Difficult: Remember that passing functions as arguments is normally implemented by passing a pointer to a function. In an RPC interface, function passing is almost impossible. Describe the circumstances, where passing a function as an argument might work in the RPC.

A6.Difficult: Passing of the function as an argument might work in RPC if we have homogenous hardware, position-independent code, and no dependencies on global variables.

Q7.Challenging: In RPC, client/server binding specifies how the relationship between a remote procedure and the calling program will be established. A binding is formed when two applications have made a logical connection and are prepared to exchange commands and data. Define persistent and non-persistent biding and give examples of when it is advantageous to use each one.

A7.Challenging: With persistent binding, a connection is set up for a remote procedure call and sustained after the call returns. The connection is then used for future RPCs. Non-persistent binding means that a logical connection is established between the two processes at the time of the remote procedure call and that as soon as the values are returned, the connection is dismantled. When the remote procedure is called infrequently, it makes sense to use non-persistent binding to save resourced needed to maintain the state information on both ends of the communication channel. If the application makes many repeated calls to remote procedures, persistent binding will save the overhead of establishing connection each time.

Q8.Challenging: Certain transport layer protocols are based on packet blast protocol. The idea is that some fixed number of packets is transmitted – a packet blast is sent and acknowledged as one unit. This is an inefficient method for large packet transfers as the entire blast has to be retransmitted in the case of a single packet failure. Suggest an alternative connection-oriented protocol suitable for large packet transfers. Also, suggest enhancements, if any, to the protocol when used in full duplex communication.

A8.Challenging: In a connection oriented protocol, in most cases, the packets are guaranteed to be delivered in order at the receiver. For large packet transfers, a sliding window protocol is ideal as the sender does not have to retransmit all the packets in case of a dropped packet.

The basic idea of sliding window protocol is that both the sender and receiver keep a “window” of acknowledgements sent and received. The sender keeps track of expected acknowledgement while the receiver keeps track of the next expected frame. The receiver sends an acknowledgement indicating the last received frame and the next expected frame. On receiving the acknowledgement, the sender advances the window and transmits the next set of packets.

When used in full duplex communication, the receiver restrains itself from sending an acknowledgement immediately. The acknowledgement can be attached to the header of the next outgoing frame. This eliminates the need to transmit control frames whose sole purpose is to acknowledge the receipt of frames. This technique of sending acknowledgement along with the data frame is called piggybacking.