Title :
Average latency and loss probability analysis of Crosspoint Queued crossbar switches
Author :
Radonjic, Milutin ; Radusinovic, Igor
Author_Institution :
Fac. of Electr. Eng., Univ. of Montenegro, Podgorica, Montenegro
Abstract :
Crossbar switch fabric is one of the most popular architectures for design of high-speed packet switches. Recently, it has been actualized crossbar architecture with buffers only in crosspoints. Such architecture has not problem with implementation of large buffers in crosspoints, any more. Performance analysis of Crosspoint Queued switch for several work conserving scheduling algorithms is presented in this paper. Since there are detailed results of throughput analysis presented in the available literature, in this paper we extended previous research with analysis of average cell latency and loss probability. Maximal buffer occupancy during the simulations is also observed so we can estimate and compare memory requirements of Crosspoint Queued switch for analyzed scheduling algorithms. It is shown that Longest Queue First has lowest memory requirements among simulated algorithms. Also, we proved results from literature that for longer buffers it becomes less important which algorithm is used. Due to easier hardware implementation and absence of starvation, Frame-based Round Robin Matching algorithm becomes most serious candidate among round robin based algorithms.
Keywords :
buffer circuits; queueing theory; statistical analysis; telecommunication switching; average cell latency; crosspoint queued crossbar switche; frame-based Round Robin matching algorithm; high-speed packet switch; longest queue first; loss probability analysis; maximal buffer occupancy; simulated algorithm; Algorithm design and analysis; Fabrics; Load modeling; Memory management; Microprocessors; Scheduling algorithm; cell latency; crossbar switch; crosspoint queue; loss probability; queue length;
Conference_Titel :
ELMAR, 2010 PROCEEDINGS
Conference_Location :
Zadar
Print_ISBN :
978-1-4244-6371-8
Electronic_ISBN :
1334-2630