SATA vs. PATA: the reality of Serial and Parallel ATA. (Serial ATA).

Examining Serial ATA's (SATA) comprehensive set of features illustrates how SATA addresses the shortcomings of Parallel ATA and the reasons for SATA as an ideal choice for deployment in enterprise-level and external RAID subsystems.

Hot-Plug Capability

Parallel ATA is handicapped by its inability to support hot-plug drives. Any Parallel ATA-based RAID solution requires the host bus into which the degraded array's RAID controller is plugged to be powered off before the failed drive can be replaced. The alternative is to continue using the degraded array that offers reduced throughput capability. Clearly, neither option is acceptable in an enterprise-level environment where 24x7 availability and high performance are important requirements. It is easy to see why SCSI and Fibre Channel that support hot-plug drives have not yielded any ground to Parallel ATA in the enterprise space in spite of ATA's cost advantage. With its support for hot-plug drives, SATA remedies this deficiency.

Decreased Width and Increased Length of Cables

Parallel ATA's ATA/ATAPI-4 standard improved signal integrity by introducing the 80-conductor cable with 40 conductive elements serving as grounds to reduce cross talk between adjacent signal lines. Though the increase in the number of conductive elements did not increase cable width over the existing 40conductor cables due to the use of thinner gauge wires, the width of these cables nevertheless impeded airflow necessary to cool a server and constrained chassis design. Furthermore, Parallel ATA continued to limit cable length to 18 inches that prevented efficient routing of its cables within chassis to reduce clutter and improve accessibility to components in the system.

SATA addresses the limitations of Parallel ATA by more than doubling cable length to 1 meter and using data cables comprised of only seven conductors--a pair of differential signal lines for transmitting and another pair for receiving, and a ground between and at each end of the transmission and reception pairs. These thin, flexible cables (with connectors only 8mm wide) can be conveniently routed to multiple drives with a very small footprint, albeit with some constraints on its bend radius. This feature is highly attractive to servers using internal RAID with high drive density.

The small number of conductors incorporated by SATA for data transmission also facilitates the deployment of backplanes for external RAID. A backplane is a physical board that is typically integrated to the backend of an enclosure. Embedded on it are multiple drive connectors connected to a plugged-in central controller via conductors etched onto the board. Note that, the central controller's interface to the host may use any protocol such as SCSI, Fibre Channel or iSCSI. Backplanes permit drives to be snapped in and mated to a connector blindly. However, due to the signal attenuation experienced with FR4 (material) based traces, the maximum length of a conductor etched between the central controller and a SATA drive connector is limited to 18 inches. Although, this limit ostensibly restricts the form factor of backplanes and enclosures for SATA, the fact is; standard racks are 19 inches wide and the maximum conductor length for SATA should permit all drive connectors to be reached from a judiciously located central controller within a 1U-3U enclosure with that width.

Increased Bandwidth

The first generation of SATA offers a bandwidth of 1.5Gbps using 10-bit signaling per byte of data i.e., equivalent to a data bandwidth of 150MB/sec. The SATA roadmap envisions two succeeding generations--SATA II and III--to be completed at approximately three-year intervals with each generation doubling the bandwidth over the preceding one. (Note that, SATA II has two phases, with phase I having been released on October 18, 2002). However, the bandwidth of 150MB/sec may not be immediately realized with native Parallel ATA drives that use dongles, i.e. serial-to-parallel adapters to offer a SATA interface. Nevertheless, SATA's bandwidth is designed to scale with that of the mature high-end interfaces such as Fibre Channel and iSCSI. Fibre Channel and iSCSI are well positioned to reach the 10Gbps bandwidth plateau within the next three to four years and SATA will not be far behind with an offering of 6Gbps in that approximate timeline. Conceivably, SATA's higher order bandwidths may influence host connectivity to entry-level external RAID sub-systems to be based on SATA itself.

Enhanced integrity of Connectivity

Ultra DMA introduced CRC-based error detection in data packets as part of the ATA-3 standard. However, no parallel ATA standard offers error detection in command or status packets. Even though the size and frequency of occurrence of command and status packets is small, the probability of errors occurring in them cannot be dismissed. SATA improves the overall integrity of connectivity over Parallel ATA by providing CRC error checking capability for data, command and status packets, thereby enhancing its attractiveness for use in enterprise RAID and external storage applications.

Enclosure Management

Enclosure management has been a long-standing staple of SCSI and Fibre Channel but absent in Parallel ATA. The SES (SCSI enclosure services) and SAF-TE (SCSI-accessed fault-tolerant enclosures) protocols provide a mechanism for administrators to manage the status of enclosures and their constituent components such as power supplies, drives, door locks, fans and temperature sensors. This is invaluable in the implementation and management of fault tolerant storage subsystems where critical components operating below their nominal states need to be quickly identified and repaired (or replaced). The administrator can generally define the periodicity with which an enclosure is to be polled, and the storage enclosure processor (SEP) makes suitable notifications. Serial ATA provides enclosure management using SAF-TE and SES, enhancing its viability as an interconnect protocol to large, reliable, external RAID subsystems.

Command Queuing and Reordering 

Command queuing allows multiple requests to be issued concurrently to one or more devices on a bus and permits those devices to reorder those requests prior to their execution to reduce latency and enhance throughput. This feature was introduced to Parallel ATA and SCSI by the ATM ATAPI-4 and SCSI-2 specifications respectively. While this feature is ubiquitous in SCSI, economic constraints have limited its proliferation in the ATA space. Parallel ATA devices are cost-sensitive, and command queuing comes at a price. Manufacturers have to incur the cost of developing firmware and integrating larger memory modules to their devices to implement this feature. Secondly, ATA is largely confined to the desktop space without emphasis on performance.

Decreased Operating Voltage

Processor cores are migrating towards lower voltages for several reasons. A lower voltage allows faster signal ramping that is crucial to enhancing speed and reducing heat dissipation on the processor. High-end processors typically have core voltages below 2V. However, to remain interoperable with other chips on the system's motherboard, they generally employ a split-rail architecture with an external input voltage of 3.3V. At the time this document was written, the most current ATA/ATAPI-6 standard specified a DC supply voltage of 3.3V for parallel ATA drivers and receivers for modes greater than 4. However, the signaling for all parallel ATA standards prior to ATA/ATAPI-6 were based on a 5V supply. Hence, to ensure compatibility with solutions that are based on standards preceding ATA/ATAPI-6, modern processors have to be 5V tolerant. However, the difficulty in achieving backward compatibility increases as processors progressively evolve with lower operating voltages. SATA solves this problem by specifying a peak-to-peak operating voltage of 500mV. SATA is clearly designed for interoperability with current and future processors.

Legacy Support

An attractive feature of SATA conducive to its adoption is its compatibility with existing Parallel ATA software. Any operating system or application that supports a given set of Parallel ATA devices can support any device in that set available with a SATA interface. Another feature is its interoperability with legacy Parallel ATA drives, achieved either by the use of integrated chipsets with co-existing parallel and serial ATA channels or the use of dongles.

Proliferation

SATA is being adopted by all major Parallel ATA device vendors (though the adoption by optical drive vendors is slower than anticipated.) Hence, the choice of devices available to users of SATA will be fairly extensive, and the cost of such devices is expected to be no higher than Parallel ATA.

Point-to-point vs. Bus Architecture

An advantage of SATA's point-to-point architecture over bus architectures (such as SCSI) is that a unique cable provides connectivity to each drive. A cable failure to a SATA-bases array is equivalent to a drive failure that can be tolerated by the use of suitable RAID. In contrast, the failure of a multi-target bus to an array that does not support as many drive failures as there are drives on the failed cable is catastrophic.

www.adaptec.com

Sanjeeb Nanda is marketing product engineer at Adaptec (Milpitas, Calif.)