AN OPTIMIZED SERVERS' CONTROL AND CONGESTION AVOIDANCE MODEL FOR QUEUE NETWORK

Abstract

ABSTRACT

Researches on optimal control of servers in queue networks are enormous in literature but none of these had considered congestion control and servers' optimality concurrently. Since congestion control is a major factor in queue network management, the need to maintain a balance between servers' optimality and congestion control had become necessary. This study proposed a Fuzzy-Treap Based Servers' Optimal Control System (FTBSOCS) which is a dual model comprising of a fuzzy-based system and a Treap-based system implemented to minimize cost of servers' usage and customers’ losses due to congestion. Fuzzy logic was adopted to ensure optimal usage of available servers by applying a fuzzy rule-based method. This method was used to derive a fuzzified decision index which determined servers' deployment pattern. The Treap-based system was used to prevent network congestion by ensuring that customers arriving the queue network when it is saturated are managed to avoid dropping or starvation by transmitting them to a tree manager, from where they are re-transmitted for service upon the availability of an idle server. The FTBSOCS was benchmarked with Adaptive-Network-Based Fuzzy Inference System (ANFIS), ANFIS-M/M/α and ANFIS-Treap. OMNeT++ was used as the simulation framework while dataset were randomly generated which served as input to the methods. With the inclusion of fuzzy logic in FTBSOCS, results indicated that average servers’ deployment was 21, 19, 15 and 16 for ANFIS, ANFIS-Treap, FTBSOCS and ANFIS-M/M/α respectively when there were 12K, 24K, 36K, 48K and 60K customers in the network. This implied that the inclusion of fuzzy logic in FTBSOCS had minimized servers’ deployment. The average of percentage utilization of available servers was 21.6%, 18.4%, 17.2% and 15.3% for FTBSOCS, ANFIS, ANFIS-M/M/α and ANFIS-Treap respectively with 12K, 24K, 36K, 48K and 60K customers in the network, indicating that FTBSOCS optimizes servers’ usage than the other methods. The average network throughput was 63.7Mbps, 63.4Mbps, 59.2 Mbps and 52.6 Mbps for ANFIS-M/M/α, FTBSOCS, ANFIS-Treap and ANFIS respectively with 12K, 24K, 36K, 48K and 60K customers in the network. This shows that ANFIS-M/M/α had a comparative performance of < 0.5% over FTBSOCS while a comparative performance of 2.9% and 4.2% existed for FTBSOCS over ANFIS-Treap and ANFIS respectively. The average number of customers dropped was approximately 204kb, 185kb, 165kb and 17kb for ANFIS, ANFIS-Treap, ANFIS-M/M/α and FTBSOCS respectively with 12K, 24K, 36K and 48K customers in the network indicating that FTBSOCS had < 2% of the overall number of customers’ dropped. The average propagation delay was 29.5, 26.3, 13.4 and 13.3 nanoseconds for ANFIS-M/M/α, ANFIS-Treap, FTBSOCS and ANFIS respectively with 12K, 24K, 36K, 48K and 60K customers in the network. This indicated that FTBSOCS had a slightly worse performance of < 0.3% to ANFIS while FTBSOCS had a comparative performance of about 10.8% and 13.4% over ANFISM/M/α and ANFIS-Treap respectively. With these results, it was concluded that FTBSOCS was more optimal in the control of servers and congestion in queue network.