CSE 228 Additional Notes on Handing Packet-frame losses in ATM Networks

Issues in Providing Performance Guarantees (e.g., loss rate) at ATM packet level vs. video frame level

QoS is a common feature in todays networks which guarantees a pre-defined quality in the service offered by the network. This is usually quantified by packet loss, throughput and latency guarantees which may be critical to many applications esp multimedia type applications.

Asynchronous Transfer mode (ATM) and other fast packet switched networks require mechanisms for providing quality of service guarantees.

QoS requirements of multimedia applications include bandwidth, delay and loss guarantees.

Effcient utilization of network resources necessitate statistical Qos management policies
existing approaches: statistical guarantees in terms of transport units. ie in terms of packets lost or packets transfer rate or time taken to transfer a packet

But multimedia applciations need guarantees at a higher level : viz in terms of media units. Thus a QoS of 1% packet loss does not translate into a Qos of 1% frame loss. This is because a frame which may break down into smaller packets may have packets depending on each other. Frame loss is inevitable if a packet of a MPEG Video frame are lost: the loss of any packet will make the later packets of the frame ineffective since the blocks are dependent on each other due to differential encoding of the dc values. Thus when 1 packet of 100 is dropped thus guaranteeing a loss of 1% it may as well mean 100% frame loss if the first packet of the frame is lost resulting in wastage of the remaning 99 packets of the frame. In such a situation, if the network has to guarantee 1% error at frame level, it will have to provide 1/100% = 0.01% error at packet level, since 100 packets constitute one frame, and loss of any one packet means loss of entire frame. This can be very expensive.

A better scheme is as follows: If the Switch has intelligent software to discard all the remaining packets of a frame whose one packet has been lost earlier, then it frees up wasted bandwidth which otherwise would have been used to transmit the useless packets of that lossy frame. Hence, if the first packet of a frame is lost (in congestion at the switch), then remaining 99 packets of that frame can be discarded by the switch. Hence, if this happens for one in 100 frames, 1% of packets get discarded/lost, and 1% frames are lost. However if the packet that is lost in the first place is for example 50th packet in the frame, then the remaining 49 packets (of a total of 100 packets) can also be discarded, thereby half the frame gets discarded (first half is already transmitted and hence cannot be discarded by switch - but is later useless when it reaches the destination since the second half has been discarded). Since 50 packets of one frame (out of 100 frames, each 100 packets) are discarded, it yields 0.5% loss of packets but 1% drop of frames. This scheme depends on which is the first packet within a frame that gets lost at the switch. On an average, this should perform much better than the scheme of the prevoius paragraph. To implement this scheme, we use one bit in the header of an ATM packet (ATM Packets are also referred to sometimes as cells): this bit is set to zero for all packets of the first frame, then toggles to 1 to packets of second frame, etc. This way, switches know which packets belong to a frame and so if one of the packets is lost , the switch can discard all packets of that frame.