Author: Marty Czekalski
Maxtor Corporation
Years ago, nearly all interfaces were serial. Communications, printers, and disks all used serial interfaces. Those were the days of discrete components, and reducing the component count was an important factor in keeping costs under control. As the level of integration increased, it became possible to put multiple drivers on the same piece of silicon as the logic. This made parallel interfaces both high-performance and cost-effective. As a result, parallel interfaces became the rule and serial interfaces the exception, except in long-distance communications, where the cost of the high-speed transmitter/receiver technology was offset by the savings in wire (or fibre) costs.
Interestingly, as technology has moved forward, the same advances in integration that caused parallel interfaces to gain favor are now working against them. The I/O cells are not scaling at the same rate as internal logic, and it is now easier and less costly to add sophisticated signal processing at lower cost than adding additional I/O cells needed by parallel interfaces. The result is that these high-performance serial interfaces are becoming cost effective at shorter and shorter transmission distances as the levels of integration increase. We now find ourselves switching back to serial interfaces such as USB for printers and PCI Express for PCI replacement. The next step will see this same migration in the interfaces we use to connect disk drives. The evolution of ATA and SCSI will move these interfaces to Serial ATA and Serial Attached SCSI. This article will explore some of the details driving this transition as well as the advantages of these serial topologies.
This new class of serial interfaces is breaking all the old perceptions of serial interfaces in terms of performance and ease of use. Because of the point-to-point nature of these interfaces, signal integrity issues common in parallel busses can be minimized. Additionally, the ability to utilize highly integrated, low-cost switching technologies (expanders) combined with full duplex connections and multiple pathways greatly improves system performance, far beyond what the simple comparison of bus speeds would suggest. For example, an eight-port Serial Attached SCSI host bus adapter is capable of up to 4.8 Gbytes/sec. peak transfer rates.
The illustration below compares an Ultra320 SCSI configuration with two initiators to a Serial Attached SCSI (3 Gb/sec. links) configuration with two initiators. In this example, each of the Serial Attached SCSI initiators is connected to an expander with a wide port consisting of two links, and the expanders are in turn connected to each other with a wide port of two links. The workload used is a burst of traffic consisting of 64KB transfers occurring closely spaced in time. It can be seen that in the case of Ultra320 SCSI, the bus becomes the bottleneck and transfers have to wait to gain access to the bus, resulting in long latencies and system congestion. In the case of Serial Attached SCSI, all the transfers occur within a short time after data is available for transfer. The combination of wide ports and full duplex operation yields a system with minimal latency that still has bandwidth available for increased workloads.
In summary, today’s serial bus technology represents a major advance in the ability to construct highly robust, high-performance storage subsystems. The advances in silicon technology have led us full circle in the way we think about and design interfaces. Connection-based switching schemes and high-performance serial interfaces that were once only considered for long-haul carrier-class applications can now be cost-effectively integrated into a portion of a silicon die. We had become accustomed to the intuitive concept that parallel interfaces provided better performance; however, advances in technology have proved this assertion incorrect. Additionally, we find that these serial technologies provide levels of scalability and flexibility that are not possible with parallel busses.
