Switched fabric architectures are changing the way designers look at multiprocessor systems in the same way that air travel changed the common understanding of time and distance. Not too many years ago, travel between Asia and the United States consumed weeks, but jet transportation reduced that time to a day or less, and made Tokyo seem close to New York.
Likewise, switched fabrics change the relative distance of distributed system resources such as remote memory from distant islands reachable only after long delays, to next-door neighbors that can be accessed almost instantaneously. Because of switched fabrics, designers can pay far less attention to which system elements are remote or local, because high-speed communication paths make everything virtually local.
However, one issue remains: just as the longest part of a trip to Manhattan is often the taxi ride in from the airport, switched fabrics frequently terminate near to, but not at, the processors they support. In telecommunications, this has been called the "last mile" problem. Computing system designers now face a similar "last centimeter" challenge of overcoming delays and bottlenecks that occur when highspeed communications are choked by the old, slow buses that link the switched fabric to the processor.
View Entire Paper | Previous Page | White Papers Search
If you found this page useful, bookmark and share it on: