Files in this item
|(no description provided)|
|Title:||Nonblocking packet switching with shift-register rings|
|Author(s):||Murakami, Gary James|
|Doctoral Committee Chair(s):||Campbell, Roy H.|
|Department / Program:||Computer Science|
|Degree Granting Institution:||University of Illinois at Urbana-Champaign|
|Abstract:||This research investigates packet switching with gigabit-per-second ports for integrated broadband services. The Pulsar switch design is based on a fast word-parallel shift-register ring, and it is a platform for studying the fundamentals of packet switching.
With simple First-Come-First-Served (FCFS) input queues, the phenomenon of Head-Of-Line (HOL) blocking lowers throughput. Simulation and analysis of several design alternatives illuminate the problem. A spectrum of queue configurations is constructed from the preliminary studies. Results confirm that non-blocking throughput can be reached with output queueing or with non-FCFS input queueing. Per-destination subqueues at each input port achieve non-blocking throughput with minimal queue memory bandwidth.
With a trivial queue service discipline, a flood of best-effort data traffic can obstruct bandwidth-sensitive video traffic. The Pulsar bandwidth accounting algorithm adjusts the priority of each stream based on whether it is below or above its allocation. A multiple-token medium-access-control mechanism implements prioritized round-robin service, and it can be extended to order traffic classes sensitive to delay or jitter. These developments form a service discipline for distributed queues that guarantees allocated bandwidth.
The Pulsar switch can be implemented easily with existing technology, and it compares favorably with other high-throughput switch designs. The design can be applied to both network switches and computer backplanes which require low latency and feedback control paths.
The research covers the fundamentals of high-speed packet switching including queue configurations for non-blocking throughput, and queue service discipline and medium-access control for allocated bandwidth. The research breaks new ground by combining non-blocking throughput and allocated bandwidth into a switch design for integrated broadband services.
|Rights Information:||Copyright 1991 Murakami, Gary James|
|Date Available in IDEALS:||2011-05-07|
|Identifier in Online Catalog:||AAI9210928|