Multiplexing Spacer Outputs on Cell Emissions zyx Guven Mercankosk, Tim Moors;, Antonio Cantoni Australian Telecommunications Research Institute GPO Box U1987, Perth WA 6001, Australia e-mail: guven, tim, zyxwv cantoni@atri.curtin.edu.au Abstract zyxwvutsr In zyxwvutsrq A T ' networks, one effective way to provide service guarantees is to shape individual trajic streams at various points in the network. In this paper, we present a method for multiplexing a large number zyxwvutsrqp of single-stream spacer outputs on cell emissions rather than on cell arrivals. The method not only approximates FCFS multiplexing of such outputs but also better suits bursty arrivals. zyxwvuts A number of performance related issues of the proposed method are identified and examined. 1 Introduction The ITU zyxwvutsrqp [ 11 and zyxwvuts A T ' Form zyxwvutsr [2] specify traffic shaping as an optional function for congestion control in protecting network resources. It forms an effective way for ATM net- works to guarantee an agreed QuaEity Of Service (QOS) for users. The logical functionality of a spacer device is illustrat- ed in Figure 1 as a reference. Cells belonging to each of the N streams are spaced in time by a dedicated single- stream spacer such that the time between successive emis- sions is at least a specified minimum and equal to that minimum whenever there is a cell in its input queue wait- ing to be emitted. Each emission instant according to the above spacing algorithm depends on the arrival pattern and is referred to as a Theoretical Emission Time (TET). Cells emitted from spacers are then multiplexed in a First emitted (Come) First Served (FCFS) manner onto the transmission system. Note that dedicating one physical spacer per stream is not practical, especially when many streams are to be spaced. The known solutions [31[41[5] to the spacing problem exploit the fact that the TET of each cell can be calculated as the cell arrives. This allows the spacer to determine when it can emit another cell from the corresponding stream. This emission could be scheduled in a calendar of evenits. It is likely that the calendar times would be quan- tized. When scheduling an emission, the targeted entry of the caleiidar might already contain an event because of a coincidental emission. The solutions mentioned above dif- fer from each other in the way that they resolve the con- tention for a calendar entry. In [3], the emission to be scheduled is placed into the closest forward empty calen- dar entry. The disadvantage of using the closest entry is the time taken to search for this entry which introduces a variable workload per arrival. The solution given in 141 avoids this problem but requires dedicated hardware se- quencer. Note that both solutions use the calendar as a list of trmmission schedules. Alternatively, the cells that are scheduled to be emitted on the same calendar entry may be kept as a linked list. As the emissions become due, the correspondinglist is appended at the end of another l i e d list representingthe transmission queue. This approachnot only eliminates the problem of variable workload per ar- rival prtxessing but also avoids the need for a dedicated hardware sequencer. A similar method has been reported in [5]. Figure 1 A reference spacer device transmission The starting point for the solutions mentioned above is the assumption that zyxw the arrival rate into the spacing device is bcuntied by the speed of incoming links which might be accurate for intermediate systems. However, arrivals can 1 b.3.1 0743-166W95 $04.00 0 1995 IEEE 49