Random Access Protocols

In random access schemes, all nodes can access a channel at any time. A classic example of random access scheme is Slotted Aloha (SA). In SA, time is divided into equal length slots and a node can access the channel at the beginning of each slot. I-SA is a generalization of single-channel SA to the multi-channel WDM network.

I-SA [84] allows a data packet to be transmitted on the destination‘s home channel in the immediate slot following packet generation. Each node has a single queue of variable capacity to buffer packets generated when the transmitter is busy. The transmitter always attempts transmitting the packet at the head of the queue. Collision occurs when more than one node transmits on the same channel. Successful packet transmissions cannot be sensed by the source node since the receiver of the source node is tuned to its own home channel. Acknowledgments have to be sent by the destination to the sender to inform about successful reception.

If a collision occurs during packet transmission, the transmitter waits for a geometrically distributed number of slots before retransmitting. The transmitter decides to retransmit in subsequent slots with backoff probability denoted by



. SA based protocols have a potential

instability problem due to high collisions under heavy traffic. Stabilizing mechanisms have been studied for SA to eliminate network collapse and provide a steady throughput [81].

Acknowledgments required in I-SA make the protocol potentially variable to propaga- tion delay. The following paragraphs discuss two different acknowledgment schemes for I-SA.

Extended Acknowledgments: A packet slot in I-SA is composed of two phases with this

approach referred as ESS [93]: data transmission phase and acknowledgment subslots. The acknowledgment subslot includes destination node transmitter tuning time and propagation delay for the acknowledgment. Nodes do not transmit data packets in the acknowledgment subslot to avoid collisions. If

M > C

, time multiplexing may be used to ensure collision- free acknowledgment transmission. Nodes resume data packet transmission at the end of the acknowledgment phase. This assumes that the acknowledgment information is small compared to the data packet size.

One I-SA data slot equals (2

a

+



+1)

T

time units, so the channel utilization per successful slot is 1

2

a

+



+1

since the initial tuning time can be overlapped. Note that the channel is being held by the source node transmitter during acknowledgment subslots. Since data transmission starts every 2

a

+



+1 slots, there is an initial synchronization delay of(2

a

+



+1)

=

2 which may degrade performance when



or

a

is significant.

I-SA was first considered for a WDM-based optical backplane to support communication in a multiprocessor environment [94]. Propagation delay sensitivity was not initially a principal concern so acknowledgment was achieved by extending the slot to include the acknowledgment. However, in a local area environment, propagation delay is a significant concern. Also, the impact of the finite transmitter tuning time has to be considered. The protocol with ESS is very sensitive to propagation delay, protocol processing latency and tuning latency.

Explicit Acknowledgments: The acknowledgment subslots described above can be re- moved by requiring explicit acknowledgments from the destination node. The slot length is now equal to the normalized packet transmission time.

A generated data packet is stored in the buffer if the transmitter is busy. When the packet reaches the head of the queue, there is an initial wait before the transmitter decides to transmit the packet at the head of the queue. This is due to transmitter contention between the data packet and acknowledgment packets generated by the node.

Packet transmission is as follows: The data packet at the head of the queue is transmit- ted and the source node enters a timeout for the acknowledgment. The timeout period is estimated based upon

a

and



. The minimum timeout is 2

a

+



+1 which includes propa- gation delay, transmitter tuning and transmission time for both data and acknowledgment. If the acknowledgment does not arrive by the end of its timeout period, the node follows a backoff policy as described for ESS. Upon successful receipt of the acknowledgment, the node transmits the next packet in the queue. Packet transmission is attempted until the data packet is successfully transmitted.

If a packet is successfully received by a node on its home channel and the packet is destined for the node, it generates an acknowledgment packet. The size of the acknowledg- ment is the same as that of the data packet. Each acknowledgment can provide feedback up to

M=C

nodes that receive it on their home channel. This is of particular advantage when there are many more nodes than channels. It also provides a form of redundant acknowledgment since all the

M=C

nodes that receive a packet may acknowledge a data packet.

Acknowledgments are subject to collision since they are transmitted as regular data packets. A data packet is unsuccessful if either the original data packet or the subsequent acknowledgment was lost due to collision. The acknowledgment can be piggybacked on the

data packet but this is not considered here. Since there are at most

M

outstanding packets in the system, the scheme described above may be sufficient to support acknowledgments. Piggybacking plays a significant role when traffic at each node is pre-sorted on transmission channel.

The implementation details of the above two schemes and the performance analysis based on semi-markov and simulation models is presented in [93].

For a single channel network, the maximum utilization with SA for an infinite population system was analytically shown to be approximately 0.36 [81]. For a

C

channel system, theoretical throughput bound with I-SA is 0

:

36

C

. However, this was not observed in the studies mainly because of the finite population model, a single transmitter and head-of-line effect due to a single queue. Also, one slot in I-SA is equal to(2

a

+



+1)

T

where

T

is packet transmission time. When(2

a

+



) 1, channel utilization per slot is poor and is

given by 1

2

a

+



+1 .

Improved I-SA: The performance drawbacks (poor utilization and head-of-line effects)

of I-SA motivated the development of I-SA* [86, 87]. The improvements are:

.

I-SA* eliminates the head-of-line effect of a single queue by providing a set of

C

separate queues per node – one per channel

.

it eliminates the ACK subslots at the end of each packet transmission by requiring an explicit ACK from the destination nodes

.

a node can transmit a window of packets after tuning to a channel thereby reducing processing latency per packet.

Discrete-event simulation models were developed to study the performance of I-SA*. The performance of I-SA* was much better than that of I-SA as reported in [86, 87]. The

2 3 4 3 2 1 0 1 2 0 5 4 3 2 3 4 3 2 1 0 1 2 Cycle C C+1 C C-1 C-2 C-1 C-4 C-3 C-5 C-4 C+2 C+1 C C+1 C C-1 C-2 C-1

...

1 C-2 C-3 0 1 C-3 C-2 0 1 2 C-2 C-1

Channels

...

Transmitting Nodes

M-1 M-3 M-2 M-1

M-2 M-1 0

M-1

Slot 1 Slot 2 Slot 3 Slot 4 Slot M-2Slot M-1 Slot M Slot 1 Slot 2 Slot 3

Figure 3.2: Allocation map for I-TDMA and I-TDMA* for

C < M

and



=0. head-of-line effects have been eliminated due to the multiple queues. The impact of tuning latency and propagation delay have been considerably reduced due to elimination of ACK subslots and transmission of a window of packets.

The following section describes the static access protocols studied for the star network.

In document HIGH-SPEED COMMUNICATION PROTOCOLS FOR ALL-OPTICAL WAVELENGTH DIVISION MULTIPLEXED COMPUTER NETWORKS (Page 49-53)