System administrators are constantly looking for ways to save money on storage equipment while still getting the best possible performance, capacity, reliability and scalability.
Performance and capacity are fast becoming the two most critical criteria, however, as the raw volume of data and long distance transmission becomes commonplace. Today's data rates for RAID arrays typically hover in the 1.5 to 2.0 GB/s range. They are expected to increase to 6.0 GB/s in the near future.
To address these needs and to keep their technologies as competitive as possible, the two leading hard disk drive interface standardsATA and SCSIhave been evolving over the past few years. Each has its advantages. While the robustness and performance of SCSI makes it the winner in mission-critical, enterprise-level applications, there is a significant gray area where the choice is not so clear.
Drive Interface Heritages
Introduced more than 20 years ago, parallel ATA, which is also known as IDE, provides a simple low-cost storage interface standard for desktop PCs. Over roughly the same time, the SCSI interface addressed the enterprise-class market that required higher performance, reliability, and scalability. As microprocessors got faster and server applications became more complex, however, the original ATA and SCSI parallel interfaces began to hit a performance wall. As a result, both ATA and SCSI have evolved toward more flexible and capable serial technology.
In 2001, a consortium of companies with an interest in keeping ATA as a viable, lower cost interface developed Serial ATA (SATA). As the name implies, the primary architectural difference was to replace the parallel interface with a serial interface. A second iteration, SATA II, was published late in 2002 to further address the needs of the networked storage market. Its improvements addressed enclosure management, backplane signaling, and cabling as well as performance enhancements.
SATA's point-to-point topology is primarily responsible for the performance enhancements. SATA does not share the ATA bus in the old ATA master/slave topology. The SATA specification can handle a maximum of 150 MB/s per storage device. Specification developers are planning increased speed transitions for SATA over the next several years.
Serial Attached SCSI
Serial Attached SCSI (SAS) carries SCSI builds on more than 20 years of SCSI technology. SAS is point-to-point technology with expander architecture. It inherited effective and efficient features from its forbearers, including storage management, and better system interoperability, flexibility and scalability.
The first generation of Serial Attached SCSI specifies a 300 MB/s connection, with each projected new version doubling the previous data rate. Although 300 MB/s is about the same as parallel SCSI's Ultra-320, SAS is a point-to-point connectionnot a shared bus. Therefore, each device gets 300 MB/s of dedicated bandwidth and this makes it more scalable.
Another advantage of SAS is that it shares the same cabling as SATA. An SAS controller will include SATA by default. The SAS protocol exceeds SATA in several areas, including support for wide links. System administrators can aggregate SAS connections between the host and subsystem. Four channels could be combined, for example, to deliver more than 1 GB/s.
In addition, SAS defines dual-ported drives and can have both ports active. Another feature of the spec is the ability to use expanders to cascade multiple devices onto a single connection or a wide link, creating a sort of storage network with a theoretical limit of 16,256 devices.
As a result of the introduction of SAS technology, SATA has to compete with both traditional SCSI with its parallel interface and SAS. Table 1 illustrates the drive interface choices available today.
||- 133 MB/s
||- At max today
||- Wide ribbon
- 18-inch length
|- 2 drives per channel
- Master/slave relationship
- Shared bandwidth among drives
||- 150 MB/s
||- 600 MB/s
||- Thin, round ribbon
- 1-meter length
|- Single drive per channel
- Point-to-point connection
- Full bandwidth per drive
||- 320 MB/s
||- None planned
||- Wide, round ribbon
- 12.5 meter (LVD) length
|- Up to 15 devices per channel
||- 300 MB/s
||- 1200 MB/s
||- Thin, round ribbon
- 6-meter length
|- 128 devices
- Expanders allow up to 16,000 devices
Table 1: Parallel and serial interface options for disk drives
There are several key areas where the drive interface technologies complete or have adopted the same technology to stay competitive: point-to-point connectivity; hot-plug support; cabling; reliability; and performance.
As data rates race well into the megabytes per second range, virtually all interconnects have to transition from parallel to serial connections to avoid the bottlenecks and data-integrity errors that plague parallel interfaces at high data rates. Point-to-point connectivity delivers both performance and reliability because each port in the controller serves a single device.
Hot-plugging or hot swapping, lets technicians exchange a defective hard drive without powering the system down or rebooting. SCSI, SATA and SAS offer hot plugging. This capability is, of course, critical for enterprise-level systems because all drives have to be serviced or replaced from time to time and it is essential to manage this without data loss or system downtime.
Staggered pins for both the hard drive and drive receptacles mate the power signals in the appropriate sequences required for powering up the hot plugged device. The pins are specified to handle more the maximum specified surge current that happens when the drive is inserted.
Cabling and connector problems are one of the most common service calls for hard drives and arrays. Point-to-point connections also require quite a bit of cable inside the enclosure, which can have a negative impact on air flow, cooling, and reliability.
The basic SATA connector design is both efficient and practical. Its "L" shaped data and power connector make plug orientation obvious to the end-user to prevent incorrect mating. The connectors are engineered for hot-plugging and connect in three stages- pre-charge, ground, and power.
The SATA connector is substantially better than parallel ATA, which has a long history of problems with bent pins. It is also a significant improvement over SCSI's daisy-chain topology because if one SCSI cable fails the entire array goes down. SATA cables are also thinner than those used is SCSI, which makes cooling the enclosure easier.
Although there is no question that SCSI drives deliver more uptime than SATA drives, system administrators often look beyond the drive itself and consider system uptime. The only way to assure 24/7 availability is to configure a system that tolerates failure of any component. And the best way to achieve this is with redundant everythingfrom multiple I/O controllers and servers to multi-path cabling.
SCSI drives offer redundancy through their dual-port technology. In a RAID configuration, data and connectivity remain intact when the drive fails. SATA's single port drive must resort to RAID mirroring to achieve the same level of redundancy. Mirroring requires twice as many drives but it also tends to enhance (about double) overall system performance by letting the system access the same data in parallel from both drives. By mirroring for redundancy, SATA also moves closer to SCSI in performance.
SATA drives lag high-end SCSI drives in performance but few system integrators use high cost, best-of-breed SCSI devices in their systems. SATA's cost advantage invites vendors to use high-end SATA drives to create systems with adequate performance at a cost lower than if they used "mainstream" SCSI devices.
The most significant point about SATA is that it narrows the gap between ATA and SCSI performance, while retaining the traditional ATA price advantage. Table 2 shows some key performance metrics.
||Enterprise SCSI (SAS)
|Latency + Seek Time
||13 ms @ 7200 rpm
||5.7ms @ 15,000 rpm
|Typical I/Os per second per drive (no RV)
|Typical I/Os per second per drive (10 rad/s)
|Typical I/Os per second per drive (20 rad/s)
|Mean Time Between Failure (MTBF)
Table 2: Performance metrics
Which is Best?
Pricing, performance enhancements, and new features have made SATA an attractive option for specific applications that were not suitable for early generation drives based on ATA technology. But SCSI still ranks as the best choice for the majority of mission-critical, enterprise-level applications for a number of reasons.
SATA's primary benefit is that it is less expensive than SCSI. It gains this price advantage in the drives themselves and in the controllers and the cables that support the drive. For budget conscious system administrators building a RAID array, the cost benefit can be quite seductive. A 200 GB, 7,200 rpm SATA drive, for example, costs about $175. For the same money, the system administrator can purchase a 36 GB, 10,000 rpm enterprise-class SCSI drive.
When properly configured and running the right kind of application, SATA RAID arrays compete on performance as well, delivering between 90% and 95% of the speed of a comparable SCSI array. In standalone configurations, SATA drives come very close to SCIS drives in terms of performance.
There are fewer management issues with SATA arrays. Both data and power connectors use thinner cables than SCSI drives. This means they are easier to move around inside enclosures. Thinner cabling also provides less restricted air flow inside the enclosure, which results in fewer overheating problems and potentially shorter component lifetimes.
On the other hand, in order to maintain its price advantage, SATA drives are not built to the exacting standards as enterprise SCSI drives. In the intensive use scenarios enterprise-class SCSI drives are still the winner. They are much more capable, for example, of easily handling massive amounts of data and they tend to be less subject to mechanical failures and surface defects. SATA drives seldom come with warranties of more than one to three years. Many system administrators have come to the conclusion that they are best employed as desktop-class drives.
SCSI also offers a superior command set. It uses a method known as command queuing to optimize data. Command queuing lets the controller execute requests for data in the order that will provide near optimal performance. A low-level method for dealing with multiple simultaneous requests such as command queuing must be implemented at the hardware level. Disk controllers can present serious bottlenecks in RAID arrays and servers, where dozens or even hundreds of users are making nearly simultaneous requests at once. SATA has a means of dealing with nearly simultaneous requests too but it is simply not as efficient.
In SATA systems, the CPU manages data flowwhich is an unfortunate but unavoidable inheritance from ATA technology. CPU loading that results from the SATA implementation is far less than that required by ATA/IDE standards but CPU power and bus bandwidth are more effectively used for other functions. By offloading data flow management to dedicated hardware in the controller, SCSI gains a critical advantage in overall data throughput.
System administrators should also be aware that for all the advantages of SATA connectors, the drives won't work with conventional power connectors for disk drives. A dedicated power supply is used and it requires dedicated power connectors or a converter. Although the cost is about $10 for each connector it does require another inventory item and adds to the cost.
In most instances, SATA is best employed for single-disk servers and desktop and workstation configurations that benefit from disk-striping setups as are required by multimedia editing stations.
Why Not Both?
The SAS system followed SATA chronologically and its architecture has been cleverly designed to support the SATA protocol. In other words, the SAS controller can recognize SATA drives. Interoperability between the two technologies accomplishes complementary goals for the competing SCSI and SATA camps. SAS systems can take advantage of SATA's price advantages and SATA drives can find new markets in enterprise applications.
First, here are two of the main differences that have to be taken into account with a hybrid system. SATA responds to a single initiator; SCSI responds to multi-initiators. SAS is a full-duplex technology; SATA is half duplex. But SAS can be dual-ported to SATA's half-duplex and single-port capability.
The expander architecture included in the most recent SAS specification means it can scale to as many as 200 devices. The SATA II Port Multiplier specification allows up to 16 SATA devices on a single SATA host port.
SAS and SATA drives can reside side by side in the same enclosure and the system controller talks to each in its own language and coordinates their activities and performance. This gives the user a high level of flexibility in configuring a storage system to provide exactly what the application requiresat the least cost.
About the Author
Contributing writer Jack Shandle is a former chief editor of both Electronic Design magazine and ChipCenter.com. He holds a BSEE degree and has written hundreds of articles on all aspects of the electronics OEM industry. Jack is president of e-ContentWorks, a consultancy that creates high-value content for publishers, eOEM corporations, and industry associations. His email address is firstname.lastname@example.org