Serial ATA: opening new markets for ATA RAID. (Serial ATA).

The enterprise data center has long been the domain of SCSI and Fibre Channel storage technology and components. Nearly two decades of technology has produced a solution set targeted to enterprise data centers with requirements for very highly available data.

ATA technology, on the other hand, has been the domain of the desktop and home markets. Attempts by various vendors to move ATA technology into the server and enterprise domain have not been successful due to real and perceived technical shortcomings of ATA drives and controllers.

A new technology is now emerging that has the potential to disrupt the long-standing polarity of these two technologies. This new technology, called Serial ATA (SATA), merges ATA disk drive technology with serial-cable technology. Seems pretty simple, so why is it being hailed as the basis for a revolution in the way we deploy storage? To answer that question, we have to look a bit more closely at the current storage technologies, and their strengths and weaknesses.

SATA Vs. SCSI

SCSI drives and controller technology was developed for server-class, 24x7, high-availability systems. The cable is long enough to support external storage, and the drives are built of heavy-duty components and tested rigorously in the factory. The SCSI feature set also incorporate a number of high-end features to improve performance and reliability in multi-user, multi-threaded servers that maximize the transfer of data between the host memory and the controller, and maximize the efficiency of the drive head across the media. The cable-interconnect schemes support up to 15 drives per SCSI bus, allowing the connection of large number of drives, making SCSI an excellent platform for large storage systems.

ATA technology, on the other hand, was developed and optimized for the desktop computer. Because desktop systems typically do not run 24 hours a day, 7 days a week, and generally are not subject to heavy usage, the components used to construct ATA drives are not as robust as SCSI. However, the cost sensitivity of the desktop market has resulted in engineering and manufacturing techniques optimized to provide a highly reliable product for the very simple reason that a drive failure could wipe out all the profit in the sale of a desktop system. The ATA protocol is much simpler than SCSI, haying been designed for single-user, single-threaded operating systems, and therefore doesn't have many of the features of SCSI. Finally, the ATA cable length is very short, only 18 inches, and supports only two drives per cable, perfect for a desktop, but inadequate for large storage systems.

RAID plays an important role in storage systems. Originally invented to provide a means of insuring data availability in the event of a drive failure, RAID has become the standard way to build storage subsystems for data centers where 24x7 data availability is a requirement. RAID creates a system where the data is redundant such that the loss of a single drive does not impact the availability of the data.

It would seem then, that a RAID system using ATA drives would be an obvious solution for high-capacity, inexpensive storage systems--the drives are cheap and RAID "hides" the drive failures--so who cares if the ATA drives are not quite as fast or reliable as SCSI? I can still get a lot of reliable storage for less money... why hasn't it caught on?

There are a few factors that have kept ATA RAID from seriously challenging SCSI: The 18-inch cable limitation of ATA makes it difficult to connect large storage banks of ATA drives; the limit of two drives per cable makes it hard to build large capacity storage subsystems, where most controllers can attain reasonable performance with a maximum of four drives; SCSI drives and controllers have features that lead to increased performance that ATA drives do not have; early attempts at ATA RAID were generally poorly implemented, causing those who tried it to have a negative experience, establishing a reputation for unreliability for ATA RAID.

Recently, the quality and feature set of ATA RAID offered by various vendors has improved over what was available just two years ago. Many ATA RAID vendors now offer a SCSI device driver model to obtain the benefits of improved host to controller data transfers, and the ATA RAID controller feature set has improved to offer features like Dynamic Sector Repair, Command Reordering, Command Read-back, and Media Scan to provide a RAID feature set that compares very favorably with SCSI RAID. These new controllers virtually eliminate the difference in the RAID feature set between SCSI and ATA.

The benefit is obvious: with a RAID system built using ATA drives and one of these new controllers, the file server thinks it's interfacing to a SCSI device, and derives all the benefits of SCSI RAID, but at a much lower cost.

SATA, as mentioned earlier, merges a serial signaling technology to ATA drives. This is important because it solves a number of problems affecting the use of ATA drives in large storage subsystems: The cable is small and flexible, allowing for easy routing and better system cooling; the cable is longer by up to 1 meter, opening up the possibility of connecting drives outside of the server; the cable uses low-voltage signaling, allowing higher bandwidth with inexpensive components, thereby preserving the price point of ATA drives and simultaneously providing the foundation for evolutionary increases in bandwidth over time; low voltage signaling also eliminates the requirement for +5VDC, enabling less-expensive power supplies, motherboards, and drives; serial signaling techniques are used, aggregating command, address, and data signals on only four wire pairs, significantly reducing high-speed signal integrity issues, allowing for much-higher bandwidth.

In addition to all of the benefits of the cable and signaling techniques, there is an additional feature of SATA that tends to get overlooked: for the first time, the connection between the drive and the controller is point-to-point instead of a bus. For every drive, there is a dedicated cable that connects it to the controller. This will dramatically change the way storage is configured and deployed because a point-to-point connection topology enables the use of switched controllers that will be able to extract much higher performance from ATA drives.

Why? Because a switched architecture enables concurrent access to all the drives, allowing for performance aggregation not possible under a bus-based architecture. Further, a point-to-point, switched architecture dramatically increases drive isolation, thereby minimizing cascading error effects that can render a bus-based system inoperable.

Bottom line: A SATA RAID controller implemented using a switched architecture will have very-high-performance, low latency, great fault isolation, may be obtained at a lower cost, and will have all benefits of SCSI RAID.

Reaping the Benefits of Redundant Storage

So, let's get back to the issue at hand: What does the availability of SATA drives and controllers mean to SCSI?

It depends on whom you ask, but most people do not believe SATA is a direct competitor for SCSI, for many of the reasons listed above: SCSI is faster and more reliable, and is the incumbent in the data center where it enjoys a good reputation.

However, SATA, especially RAID SATA, creates some intriguing opportunities for inexpensive, reliable storage for classes of data that have traditionally not reaped the benefits of redundant storage. For example, if you look at a typical server and the data stored on it, you will generally find: Financial and inventory data is stored on SCSI RAID; everything else may or may not be on SCSI, but it is certainly not RAIDed--home directories, archives, video data, file servers, etc. are all subject to loss of data due to a disk crash.

This, then, is the opportunity for SATA--to protect all data with RAID. With SATA, all of your corporate and personal data can be inexpensively protected from a disk crash. This market is far, far larger than the current market for SCSI RAID. Thus, SATA will not so much compete with SCSI, as extend the benefits of RAID into new markets.

Yes, there will be areas where SATA competes head-to-head with SCSI. The current state of the economy is providing much of the impetus behind this movement--customers need the benefits of RAIDed storage, but they simply cannot afford the cost of SCSI, so they are open to alternatives. ATA, especially with the advances in ATA RAID mentioned above, is now being successfully deployed in traditionally SCSI-only applications. And, as customers try ATA RAID and become comfortable with it, they are more willing to try it in more applications.

However, there are significant markets where SATA will pioneer RAID:

Corporate files servers storing non-critical data: The low cost and good reliability of SATA RAID makes this an ideal solution for insuring all data is protected by the benefits of RAID. Home directories, email servers, and corporate file servers can all be RAIDed with SATA technology to provide better data availability at a low cost.

Tape replacement and near-line storage: If you were to compare the costs of a tape system with a system of equivalent capacity built using SATA technology, you will find that SATA storage is a very cost-competitive alternative to tape. Disk-based technologies offer two additional benefits over tape systems: They can be written at disk speeds, thereby dramatically reducing the time to backup your data, and they can be read easily, using normal file system commands, resulting in very rapid recovery and restoration of backup data.

NAS and SAN: The trend in servers is to move the storage out of the server and onto a NAS or SAN network. NAS typically exports a NFS or CIFS file system protocol over Ethernet, while SAN exports a SCSI or Fiber Channel interface and "looks" like a local drive to the server. The point here is that the interface to the network makes no requirement on the storage technology inside the storage unit. Thus, it is perfectly legitimate to build NAS and SAN storage containing SATA drives, but exporting any number of network interfaces, including NFS, CIFS, iSCSI, SCSI, or Fiber Channel. NAS and SAN vendors are already building solutions containing ATA drives. SATA will accelerate this trend because it's point-to-point connection topology makes it easier to connect large numbers of drives to a single controller and create RAID sets. The result is less expensive NAS and SAN solutions.

Data archives: Data archives elicit visions of huge vaults with miles of shelves containing tapes, with bespectacled technical "librarians" keeping track of what's stored on each tape, and checking them out to researchers for analysis. The mind boggles at the amount of work it must take to load data from multiple tapes onto a system in order to perform analysis! SATA technology can simply this significantly because disk-based technology offers all the benefits of speed for archival and retrieval; and the hot swap features of SATA enable the ability to manage SATA disk drives with robots, thereby allowing researchers to "load" and "unload" the drive or drives they want, and perform data analysis much more easily.

Video streaming: This market has a unique requirement--it cannot tolerate a disruption in the data flow or the picture you're viewing on the screen will glitch. In a typical file system application, drive latency can be tolerated because the user is more interested in getting the data than in getting it fast. In streaming video, however, there is a critical time window in which the data must be delivered or the video image becomes disrupted. The requirement, therefore, is that the latency associated with data reading and writing must be very small, and must also be consistent. Traditional bus-based storage architectures like SCSI and ATA have difficulty meeting the performance and latency requirements of video streaming. However, the switched architecture implicit in SATA's point-to-point architecture solves both problems by allowing drive aggregation for higher performance, and virtually eliminating arbitration latency. Use of RAID, particularly RAID 10, which combines mirroring and striping, provides both h igh speed and redundancy, which is ideal for this market.

Digital video: This market combines the requirements of data archival with the performance and low-latency requirements of video streaming. Digital video generates huge files that must be stored, yet be readily available for editing. This requires large amounts of fast storage that can be quickly retrieved for editing. There are now "digital VCRs" available that use disk technology instead of tape for high-speed access and ease of storage.

Another market in the digital video arena is security systems based on disk drive technology instead of video tape. The digital VCRs mentioned in the preceding paragraph are replacing VCRs in security systems. By combining disk drive technology with software, it is possible to have highly intelligent security systems that can store effective and relevant data on disk where it can be quickly searched for security events.

These markets were initially being driven by White Box VARs and System Builders looking for ways to compete with the branded storage solutions. By building systems containing ATA technology, they were able to offer storage solutions at a much lower price. As these VARs and Systems Builders became more proficient with ATA technology, they began bidding solutions on larger and more complex storage opportunities. Today, ATA drives are competently doing the job at many small to medium businesses, universities, video and security applications, and many government applications requiring huge amounts of storage.

The entry of SATA into the market comes at a time when ATA technology is gaining acceptance, and the continued high cost of SCSI, particularly in the current economy, bodes well for the acceptance of SATA as a viable, reliable storage technology. Over the next few years, SATA will encroach on the SCSI market to some degree, but the primary growth in SATA will be in markets SCSI doesn't serve well today due to prohibitive costs or excessive latency.

www.3ware.com

Mike Wentz is Senior Director of Marketing at 3ware (Mountain View, Calif.)