This preview shows page 1. Sign up to view the full content.
Unformatted text preview: usually takes place
through the shared memory. These machines are easy to program, but the number
of processors that can be usefully deployed is usually small (16 to 32 CPUs) and
limited by the bandwidth of the shared memory.
2. Distributed memory parallel computer. As shown in Figure 20.6, in this type of
parallel computer, the processors do not share memory, and each processor has its
own local memory. All communication between the processors is done by passing
messages across the network that interconnects the processors. Hence distributed
memory parallel computers are also known as message passing parallel computers.
They are more freely scalable and can have an almost unlimited (several hundred)
number of processors.
Local memory Local memory Local memory Local memory CPU CPU CPU CPU Communication network
Figure 20.6. A distributed memory parallel computer.
The speed of modern supercomputers is measured in gigaflops and teraflops. A
gigqflop is one billion (109) floating-point arithmetic operations per. second, and a
teraflop is 1012 floating-point arithmetic operations second. The PARAM 10000
supercomputer (shown in Figure 20.7) developed by the Centre for Development
of Advanced Computing (C-DAC), India in the year 1998 is a 100 gigaflops
Like mainframe systems, supercomputers also support multiprogramming and can
be accessed by multiple users at the same time. However, the main difference
between these two types of systems is that supercomputers primarily address processor-bound applications whereas mainframe systems are oriented to
input/output-bound applications. In processor-bound applications, the amount of
work that can be performed by the computer system is limited primarily by the
speed of the CPU(s), whereas in I/O-bound applications, the amount of work that
can be performed by the computer system is limited primarily by the speeds of the
I/O devices. Supercomputers are rarely used for processing input/output-intensive
applications, such as payr...
View Full Document
This document was uploaded on 04/07/2014.
- Spring '14