In cell ( a d o r packet) switching networks offering multi-category cell transportation services [1,3,4,11,13], it is highly desirable to use a single Congestion-level Indicutor (CI) for a group of queues. This reduces the amount of the $ow of network management and control information across different components and sub-components of nodal switch and. across the network, Simple CI --like the maximum of the CIS of the queues in a group --causes reduced utilization of storage resources distributed throughout the network. And hence, it gives false indication regarding availability of network resources. Also it adds high overhead especially when the number of queues to be monitored is large (i.e., 2100) . Few intelligent CIS to overcome these problems are discussed and evaluated in this paper:
1: IntroductionCongestion results from scarcity of resources. When the demand for resources exceeds (or almost equals) the amount of available resources (e.g., processing power, buffer space, transmission capacity, etc.) congestion begins to be visible, because the prespecified quality of service can not be met. The issues of nodal congestion are addressed here. If adequate transmission capacity is not available, network level congestion is initiated.Note that the traffic pattern (or profile) is an important factor which determines the effectiveness of all the congestion control/ managemenf techniques deployed in a nodal operating system or in a network. It mav be nossible to use various convestion sicnatures along with some nreventive techniaues to controUmanuge sonrestion when the traffic nrofile can not be nredicted accuratelv. The signatures may include: time-duration signature, intensity signature, and spread signature, etc. of traffic variation.
2:Three different techniques can be considered:Techniques for Aggregating Congestion Level (i) The simplest one: It takes the maximum value of current the congestion levels of all the queues (ii) The Weighted Average approach: It takes a weighted average of the congestion levels of all the queues being monitored. The pre-assigned weight of a queue is directly proportional to the pri-156 0-7803-1820-X/94 $4.00 0 1994 IEEE ority category of the packets it is serving. ONLY A GROUP OF ELIGIBLE QUEUES is MONITORED out-of a large number of queues feeding the server (e.g., a communication channel), as shown in Figure 1. A queue becomes eligible for monitoring ONLY when it crosses a pre-specified congestion level (e.g., LEVEL-1) in any LAST THREE consecutive sampling epochs. A Queue can DYNAMICALLY join and/ or leave the group of monitored QUEUES. A queue leaves the pool of 'Being Monitored' queue ONLY when its occupancy level STOPS crossing the LEVEL-2 in any LAST TWO consecutive sampling epochs. That is, a queue changes its STATE from being-monitored to not-being-monitored ONLY when it STOPS crossing LEVEL-2 in at least any LAST TWO consecutive sampling epochs. Note that the sampling interval is proportional to the service rate (or link speed). (iii) The Sigmoidal Thresholding Approa...
