by May 6, 2001 0 comments



When was the last time you heard an announcement about an increase in processor or hard disk speeds? Was it yesterday, or the day before? Advances that increase the speed of data processing and capacity for data storage are the most common in computing. What do not move forward as fast, are technologies for data acquisition and transfer between the various subsystems of a computer. With processor speeds zooming up, these could then become bottlenecks in number crunching.

Data moves between the subsystems through the bus. The bus has had an interesting evolution and now stands poised for even more exciting things. Here is the story of the bus.

When ISA and PCI were King

ISA (Industry Standard Architecture), which was released in 1981, was the first popular bus architecture. It had two versions, one each for the IBM PC/XT and PC/AT. It was the AT version, called the AT bus or ISA, which became the industry standard. But with its 16-bit data path, 8 MHz clock speed, and maximum data transfer rate of 5 MB/sec, ISA could not keep up with the faster data transfer needs that were required with the advent of video, LAN, and the GUI.

It was to meet this that in 1987 IBM introduced MCA (Micro Channel Architecture). Since MCA was not backward compatible with ISA, it did not gain acceptance. Meanwhile in 1998-89, the so-called Gang of Nine, competitors to IBM (AST, Compaq, Epson, HP, NEC, Olivetti, Tandy, WYSE, and Zenith), designed the EISA (Extended Industry Standard Architecture) to replace ISA. EISA was backward compatible with ISA and was used in 386 and 486-based PCs, but that too did not quite catch on.

The technology that did succeed, and the one that is the standard today, is PCI (Peripheral Component Interconnect), which replaced ISA as the industry standard in the early 1990s. PCI was developed by Intel, but it has become so popular that you will now find it in almost every computer, including in Macs.

Running at clock speeds of 33 MHz and yielding a throughput of 132 MB/sec, PCI can support three to five peripheral devices, which may be integrated directly onto the motherboard or added as PCI expansion cards. Another reason for PCI’s initial popularity was that it supplemented rather than replaced the existing buses. Existing ISA cards, for instance, could work along with
PCI.

New technologies

Now, because of advancements in data storage and processing technologies, PCI itself has been sitting in an unsure position. Newer technologies are vying to replace the by now ageing
PCI.

NGIO, Future I/O, and InfiniBand

Intel and Sun were the first ones to propose a successor to PCI with NGIO (Next Generation Input/Ou put). With speeds of a single NGIO connection being 1.25 or 2.5 Gb/sec, this was designed for servers, which have heavier needs for fast and reliable data transfer.

NGIO used a ‘switched fabric’ instead of the bus architecture that PCI uses. In a bus, all the components are placed along the same line and share the bandwidth available. With the ‘switched fabric’ different components plug into a switch that can connect the components to the system processor as needed. Intel had argued that in a bus system, the failure of even one transaction along the bus could affect the entire system. In a ‘switched fabric’, components were better isolated from the processors. So a failed component wouldn’t bring down the entire system.

Soon, there evolved a bus-war similar to the one between MCA and EISA when IBM, Compaq, and HP preferred a competing standard, Future I/O. This, too, was based on switched fabric and channel-based communication that bypassed the traditional I/O bus. After months of wrangling, the two factions agreed on a compromise that goes by the name of
InfiniBand.

HyperTransport

The most recent contender in the race for the bus is HyperTransport. This is a technology propounded by AMD, which promises to increase both the speed and amount of data that gets transferred between the processor and computer components. This technology is supposed to be 24 times faster than the PCI bus, and aims to offer a peak data rate of 6.4 GB/sec. 

API Networks (earlier known as Alpha Processor), which helped develop HyperTransport, has recently launched a chip, AP1011, based on HyperTransport. AMD plans to use this technology in its forthcoming Hammer family of 64-bit processors for desktops and servers. 

HyperTransport will probably be used first in networking equipment such as routers, cellphone base stations, or optical switches, and later in PCs, set-top boxes, or handheld computers. It is in the use of this technology in the PC that AMD will face competition from Intel, which is working on yet another, undisclosed bus standard for desktop computers.

Juhi Bhambal

No Comments so far

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.

<