Serial Attached SCSI emerges: profound changes ahead.

Nearly four years after work on the Serial Attached SCSI (SAS) specification began, SAS products have now entered the market. Thousands of meeting hours and specification writing, plus component and infrastructure development, have all culminated in the first round of servers, host bus adapters, disk drives, enclosures and production-capable infrastructure.

Initial Deployment of SAS

Initial SAS deployments will take on one of two evolutionary forms, the first being a simple replacement for parallel SCSI. SCSI is still the dominant enterprise drive connection scheme nearly 25 years after initial shipments and preserving this industry investment is paramount in sustaining many business critical environments. Most of the initial SAS deployments, performing as a parallel SCSI replacement, will be found in traditional Direct Attached Storage (DAS) environments located in or near the server.

Parallel SCSI Replacement

As a parallel SCSI replacement, SAS maintains legacy SCSI middleware compatibility and extends its value by improving overall system scalability (distance and addressing), improved availability, (dual-porting), improved performance (full-duplex and port aggregation) and improved reliability (point-to-point architecture).

An incremental, but incredibly significant value addition to SAS, is the ability for both enterprise-class SAS performance-oriented drives and desktop-class Serial ATA (SATA) low-cost bulk storage drives, to co-exist within the same SAS infrastructure. This dramatically augments the legacy value of SCSI by allowing architectures that can support a variety of tiered storage application needs. These deployments will leverage common building block components to span the diverse storage needs in terms of performance and cost (Table 1).

SATA Uncompromised

In the second type of early deployments, SAS will emerge as a "super-sized" version of SATA. SAS has often been called SATA without the compromises. SATA deployments in the enterprise have faced numerous obstacles. Routing SATA signaling across back planes, scaling SATA to meaningful drive counts, providing robust failover mechanisms, as well as supporting multi-initiator environments come with their own set of challenges.

SAS neutralizes a number of these challenges by providing an industry-standard method of bridging between the two technologies. Using the SATA tunneling protocol (inherent to the SAS standard), SATA connections can be managed at the endpoints of the topology, usually between a SAS expander and the SATA drive, without requiring significant routing of SATA lines across backplanes. This can greatly improve signal integrity within these SATA-based subsystem solutions and can significantly enhance overall system reliability. Additionally, SATA cabling schemes offer significant limitations in terms of providing meaningful storage expansion external to the server. Routing SATA protocol over the enterprise-capable SAS infrastructure removes these system-scaling barriers for SATA-based systems. Finally, the ability of SAS to scale beyond 16,000 connections makes it suitable for implementing systems requiring large numbers of SATA drives.

Supporting SATA drives through SAS expanders also provides a means of supporting the enterprise necessities of failover and multi-initiator when implementing SATA-based storage systems. SATA drives are primarily cost driven and as such, most new capabilities in SATA II are considered to be optional (so as not to impose further costs on the primary volume desktop market segments). SAS functionality encompasses all of these SATA enhancements and provides an interoperability framework for SATA and SATA II offerings with dissimilar feature sets. In addition, SAS RAID controllers will support RAID 6 capabilities, allowing a second drive to fail during the rebuild cycle of an initial drive failure--an especially important and significant feature for "fat" and relatively slow SATA drives.

We expect to see initial "super-sized" SATA deployments occurring in some traditional DAS-based infrastructures, but mostly into new categories of tiered storage solutions including disk-to-disk copy, near-line storage and other archiving solutions. A bulk of these deployments will occur in the emerging segments of the Storage Area Network (SAN) and Network Attached Storage (NAS) subsystem markets that have begun to embrace SATA in the enterprise.

SAS Impact on Server Architecture

These initial deployment models, while very meaningful, are just the beginning. SAS promises to make an even more profound impact on server architecture, as the technology matures and becomes more widely deployed. Some key themes that will influence how servers are implemented and deployed are discussed in the paragraphs below.

Size Matters

SAS was architected to embrace small form factor (SFF) 2.5" enterprise drives. Cable and connector form factors allow for new system capabilities that are just now beginning to be explored. These smaller drives allow for greater airflow in temperature-challenged server environments, allow for full RAID 5 or 6 implementations that can be housed in IU "rackable" form factors and allow for a dramatically large number of drives to be configured in storage racks and shelves, thereby permitting an overall reduction in computer room floor space. Furthermore, server blades are often configured with dualboot drives and the 2.5" form factor promises to greatly improve the size, cooling and rackability for this category of server blades.

External connections have also improved with SAS. InfiniBand-style connectors and the rapidly emerging high-density connectors, provide small, manageable connections between racks or subsystems without burdening the system in terms of size or cost.

Performance

An immediate contributor to storage performance is the ability of SAS to take advantage of the smaller drive sizes and greatly improved address-ability to move more heads concurrently across more drives. Some early benchmarks of initial SAS systems are demonstrating substantial performance improvements for a variety of I/O workloads. In addition, SAS' ability to aggregate ports provides very large pipes that can handle tremendous transaction processing workloads that scale by adding more drives.

Some High Performance Computing (HPC) applications have begun to look at SAS as a means of providing low-cost, low-latency, high-bandwidth connections for applications reaching beyond SCSI's use as a traditional storage interface. For little more than the cost of a dual-channel Ultra320 SCSI controller, an 8-port SAS controller operating over a pair of small connectors--about the size of two SGVA video ports--can deliver an aggregate bandwidth of 48Gb/sec. The industry has yet to fully comprehend how these low-cost, small-form-factor, high-bandwidth connections will impact server architecture moving forward.

Stitching it all together

The inherent nature of a point-to-point architecture also allows the SAS topology to move beyond the traditional role of parallel SCSI. The cost, power, size and performance make SAS a suitable way to "stitch" server blades and storage blades together. The notion of removing disk drives from server blades is not new, but most implementations call for the system to provide a boot-from-SAN capability. Numerous efforts to standardize how booting from a SAN is accomplished have been, or are being, implemented, but these solutions are far from being prevalent.

Improvements in SAS expanders to permit zoning capabilities allow for the mapping of every server blade to individual boot drive(s) contained within a physically separated SAS storage blade. This implementation leverages the proven SCSI boot model, which has been the foundation for the volume server market segment for over two decades. Relocating the SCSI boot drives from the server blades onto a shared storage blade brings a new level of efficiency to server blade architecture and further enhances the size, performance and scaling attributes in these systems. (Refer to Figure 1)

Since SAS was always more device-oriented than network-oriented, its point-to-point connection model that can be scaled with low-cost expanders, provides an effective means of clustering across a server mid-plane or near the server complex. Scaling to larger configurations necessitates complementary cross-campus networking schemes and allows for these distributed SAS clusters to scale in new and innovative ways. Software to enable these rich new architectures is already being developed.

Let the journey begin ... SAS' much anticipated market introduction is finally upon us and meaningful evolutionary deployments have begun in earnest. Additionally, the new capabilities of SAS promise to greatly impact system miniaturization, system performance, HPC computing, clustering and grid computing and are likely to be combined in some very inexpected ways that will revolutionize server architecture far beyond the realm of traditional SCSI usage. The emergence of SAS has brought a new level of excitement into storage businesses, but the more profound impacts of this technology are still to come.

Harry Mason is director, industry marketing at LSI Logic Corp., Milpitas, CA

www.lsil.com

www.scsita.org

TABLE 1. INITIAL SAS DEPLOYMENTS

CASE 1: Ultra320 SCSI Replacement  CASE 2: "Super-sized" SATA

Investment Protection              Improved Reliability
  * Legacy SCSI Compatibility        * Signal Integrity
SATA Compatibility                 Improved Scalability
  * SATA Tunneling Protocol          * Addressability
Improved Scalability               Out-of-box Expansion
  * Addressing/Cabling               * External Cabling
Improved Performance               Multi-Initiator Support
  * Bandwidth & Wide Links           * SATA Tunneling
Investment Availability            Failover Support
  * Dual-porting                     * Dual Porting
Improved Reliability               Improved Availability
  * Point-to-point                   * (RAID 6)