Author: David So, LSI Logic
Video on demand. Tivo. On-line gaming. Digital photography and videography. Video and audio editing. E-mail archives. New SEC and federal regulations (SEC 17a-4, Sarbanes-Oxley, HIPAA, etc.). The increasing need for storage capacity continues down a seemingly endless path, and the timing for Serial Attached SCSI (SAS) couldn’t have been scripted any better. SAS introduces greater storage scalability and the ability to access that storage faster than ever before.
SCSI is Good
Today’s dominant disk drive interface in the enterprise server and direct-attach storage markets is Ultra320 SCSI, a parallel SCSI speed bump that is nearing its end in innovation. As has been announced by the SCSI Trade Association, the next evolution of SCSI technology beyond Ultra320 is SAS. While Ultra320 SCSI offers many benefits over other currently available I/O interfaces and has served the market well, its physical limitations leave something to be desired as data centers clamor for even greater capacity and higher bandwidth. Parallel SCSI is limited to 15 target devices per bus, and a shared bus at that. With today’s prevalent dual-channel SCSI host bus adapters, that’s a total of 30 disk drives per HBA.
Although a low-voltage differential SCSI bus is specified to be up to 12 meters in length, a heavily loaded SCSI bus is, in actuality, much shorter. Even with SCSI bus expanders, the scalability of Ultra320 SCSI is still relatively limited, and since it’s a shared bus, the theoretical maximum throughput is 320MB/s, regardless of the number of disk drives attached on that bus. That equates to somewhat restricted scalability from the perspectives of both capacity and performance. The next evolution of SCSI needed to remove those restrictions and, at the same time, continue the (expected) progression in performance.
But SAS is Better
SAS introduces a new inflection point in I/O storage technology. Its point-to-point architecture provides dedicated links for each end device at 3Gb/s (6Gb/s full duplex). With those dedicated links, a single I/O or drive failure won’t result in a hung bus that prevents the user from accessing the other disk drives as may occur with parallel SCSI. Furthermore, SAS dual-ported disk drives allow connections to more than one host controller, eliminating single-point of failure. Also the ability to aggregate phys into wide ports adds yet another level of reliability unseen in previous SCSI generations.
In addition, SAS offers unprecedented flexibility with its support of both SAS and Serial ATA (SATA) devices. With SAS infrastructure, IT managers can specify the most appropriate drive based on the need: SAS for online, transactional, high performance and high availability; or SATA for nearline, archival, reference and low availability. Such flexibility minimizes system development and infrastructure redundancy, reducing hardware and IT management costs.
SAS is Scalable
What truly sets SAS apart from its I/O counterparts is its scalability benefits. In fact, SAS infrastructure offers scalability to the SATA market that was once unthinkable, offering potential deployment of hundreds of SAS (and SATA) disk drives in a single domain. IT managers have increasingly looked to SATA for nearline storage applications, and SAS has been a boon to SATA disk drives in the enterprise market. It is true that SATA II has introduced SATA port multipliers into the mix and that these devices certainly improve the expandability of SATA. Is that enhanced expandability sufficient for the ever-increasing demands of the enterprise market? Although port multipliers are available for 4-, 8-, and even 16-drive connectivity, they cannot be linked together, thereby limiting their expandability for larger drive topologies.
The key to SAS scalability is a device known as an expander. SAS expanders are essentially high-speed switches, each one capable of connecting as many as 128 end devices. Expanders are (or will soon be) available in port counts ranging from six to 36, and they can be cascaded together, unlike SATA port multipliers. With SAS expanders, enterprise systems may reach drive counts as high as 16,256! That certainly exceeds the drive/capacity scalability of SAS’ SCSI predecessor.
Performance scalability is equally as important to the enterprise market as drive scalability and once again SAS answers the call. As mentioned earlier, since SAS is a point-to-point topology, performance is much more scalable as opposed to a shared bus interface such as Ultra320 SCSI. With 3Gb/s of dedicated bandwidth for each target device, I/O performance scales with each additional drive added to the topology. In fact, with fewer disk drives than can be connected on a parallel SCSI bus, SAS effortlessly saturates a 133MHz PCI-X host bus, as illustrated in Figure 1 below.
Figure 1. SAS vs. Ultra320 SCSI Performance
Native x4 and x8 PCI Express-to-SAS controllers will be available from multiple vendors later this year. As shown in Figure 1, PCI Express eliminates the bottleneck at the host bus, and its 250MB/s (500MB/s full duplex) bandwidth per lane will allow SAS performance to scale well beyond the limitation of the PCI-X ceiling. In other words, PCI Express will further allow SAS to showcase its performance scalability and that it is up to the task for even the most demanding of mission-critical environments.
Conclusion
As the thirst for more storage capacity continues to grow, greater requirements are placed on the I/O interfaces to satiate that thirst. SAS was specifically engineered for that purpose. Its deployment flexibility in supporting differing classes of disk drives for both online and nearline storage makes SAS infrastructure even more compelling for the evolution of Information Lifecycle Management and the implementation of tiered storage architectures. The scalability of both performance and capacity strengthens SAS technology as the I/O interface of choice to meet the constantly increasing demands of present and future enterprise storage applications.
