UDMA vs. SCSI RAID. Which One Is Right For You?

There's a relatively simple connectivity technology rocking the storage industry. It's not flashy like IP Storage, nor does it have the bandwidth of SCSI, but it gets the job done, particularly in desktops and workstations. The technology is ATA.

ATA/UDMA drives are gaining some ground in the high-end workstation market where SCSI-based drives have long been unchallenged. They're also cropping up more frequently on low-end workstations. The main value proposition for ATA/UDMA drives is that they are very inexpensive and deliver sufficient performance for most applications.

There are fundamental differences between ATA/UDMA and SCSI drives, and their effectiveness when used in a RAID array. This article examines these differences and discusses some of the key issues you should consider if you're thinking about setting up a RAID array with these drives.

Price vs. Performance

The first advantage associated with ATA (vs. SCSI) is low cost. Traditionally high-end ATA drives are priced at one-third the cost of equivalent SCSI drives. A high-end, Ultra160 SCSI drive sells for about $500 today. A high-end ATA drive is priced at just $180.

But the price/performance lines between ATA and SCSI are starting to blur. Some disk drive manufacturers are now reported to be developing high-performance SCSI drives that will be available at ATA prices. ATA drive makers, on the other hand, are also developing more robust drives that include an increased number of SCSI-like features and functionality.

In the end, the customer will win with greater I/O speed at much lower costs than has been previously available.

Performance Differences

Perhaps the most significant difference between SCSI and ATA is that ATA drives are single I/O devices with a point-to-point protocol. A SCSI interface allows you to have up to 15 drives per bus performing tasks concurrently. With an ATA interface, you can only have up to two drives per bus and, while one drive is performing a task, the other must remain idle.

In a RAID scenario it is very important to overlap commands as much as possible in order to increase array performance. With SCSI, I/O commands can be performed across up to 15 drives concurrently. With an ATA interface, they can't. Therefore, there are some limitations to the performance you can achieve using ATA drives in a RAID array.

In workstations these limitations are less important than in a server, since the applications tend to be single-threaded. In servers, because there is typically a queue of active requests from multiple users, the ability of SCSI to satisfy those requests concurrently greatly adds to the performance achievable with the array.

Another point to consider when planning a RAID array is cable length. With UDMA/ATA there's a very short maximum allowable cable length between the controller and drives of 18 inches, meaning that ATA interfaces are limited to "inside the box" RAID applications. SCSI, on the other hand, allows cable lengths of up to 25 meters between controller and drives, providing greater flexibility for setting up external or internal RAID arrays.

A third performance differentiation is the command set. With UDMA the instruction set used with hard drives is very low level and sparse. There are only a handful of simple error codes and a very limited command set. SCSI offers a rich command set and detailed error codes.

Another point to consider is the number of devices supported by each PCI controller chip. Each ATA drive needs its own controller. This means that in an UDMA/ATA RAID array there are a lot of interfaces and a lot of point-to-point cables and controllers. In the end the savings earned from using less-expensive drives might not outweigh the total cost of setting up the RAID array.

One final point worth mentioning is that an ATA bus is not designed to support hot-plugging or even drives failing during a data transfer. This would require a much more complicated error handling code in the firmware and controller chip. It is quite common for a failed ATA device to make the bus unusable for the other drive. SCSI, on the other hand, supports hot plugging and provides BSY and RST signals, making it quick and easy to revive a bus that has become unstable.

RAID Applications

Most workstation applications are single-threaded and stall, pending the completion of an I/O. Therefore, for a workstation, a very appropriate architecture is to implement your RAID using the host CPU. This means simple interfaces and simple implementation, and it will probably outperform a microprocessor-based I/O controller you might otherwise use.

In addition, high-end workstation applications are becoming more dependent on high performance I/O. A good example is high-resolution image manipulation, which requires a large amount of virtual memory space. Because most workstations are not configured with a lot of memory, a great amount of demand is placed on the workstation's I/O. High-performance I/O can greatly improve application performance.

On low-end workstations, in I/O-intensive applications like these, ATA/UDMA drives with RAID are out-performing SCSI drives in terms of price/performance. On high-end workstations, however, SCSI remains the price/performance leader.

About Adaptec RAID

Adaptec is a pioneer in I/O technology and one of the first companies credited with making the complex world of connectivity simple and affordable with SCSI. Today Adaptec is doing the same with RAID. The company's vision is RAID on every server.

In the spirit of this initiative, the company is currently developing new UDMA RAID products, slated for introduction in mid-2001. The products will provide low-cost data protection and increased performance for workstations and servers.

Ed Rebello is the RAID Communications Manager at Adaptec, Inc. (Milpitas, CA).