Author: Dan Tanner, Founder and Principal
ProgresSmart
“Feeds and speeds” matter to engineers and are quoted by shallow analysts. Price matters to the business buyer, but only in the context of effectiveness. This brief article explains why you should care about Serial Attached SCSI (SAS).
SCSI (pronounced “skuzzy”), is the acronym for Small Computer System Interface. SCSI is a standard that had been around for years, widely used in high-performance workstations and servers. But until recently, SCSI cabling and electrical interface employed cumbersome and limiting wide parallel ribbon cables and wide connectors. Making the SCSI electrical and physical cable/connecters into slim serial parts has removed complex speed to precise cabling matching requirements and eliminated the severe configuration restrictions that had plagued parallel SCSI and which had forced the invention and employment of Fibre Channel (FC) or even Internet SCSI (iSCSI). In the bargain, SAS saves on both space and, importantly, uses far less power.
Simply put, disk subsystems based on SAS cost less than similar subsystems based on either FC or even iSCSI. And that is in addition to the space and power savings. But the cost-to-purchase saving is only the tip of the savings iceberg. Businesses have often learned to their chagrin that the bulk of the cost of a storage subsystem lies in the operational management expense. That is the case in spades where FC is concerned; storage administrators require specialized training and experienced storage administrators naturally command premium wages.
Back to basics: SCSI is the standard peripheral device interface for small computers (that is, Windows, UNIX and Linux on Intel & Mac and similar hardware). FC uses the SCSI program command structure but imposes a complex and costly-to-own and operate subsystem interconnection network that is typically implemented on fiber optic cabling. iSCSI is a clever and more economical way to use the SCSI program command structure over the Internet protocol (more about that later). SAS is SCSI. It uses the SCSI program command structure with full transparency. That means that users see no difference and need make no adjustments to programs, methods, or operations when SAS is the underlying infrastructure.
Moreover, the disk programming/command interface for inexpensive desktop systems, called Advanced Technology Attachment (ATA), is a totally included subset of the SCSI programming/command interface. And there is a Serial ATA (SATA) electrical scheme – ATA also had a need to be freed from parallel connectivity – that has been designed into SAS as a subset. This is enabling disk subsystem makers to offer products that can cost-effectively mix less expensive, lower performance SATA and high-performance server-class SAS disk drives in single-package units.
In the past, ATA was used for inexpensive non-critical systems, SCSI and/or FC was used in higher-performance critical systems, and iSCSI was sometimes used as a cost saving alternative to FC and, because it can be used over a wide area network (the Internet), for long distance remote disk subsystem connection. SAS is fully capable of being used across the performance and criticality spectrum, with its only limitation being long distance (this is what I promised you I’d come back to later). SAS is ideal for “bounded” systems – those in a single-room location. But there is no programming barrier to using presently-used software to, for example, remotely replicate files between a SAS-based disk and an iSCSI-based subsystem in a far-away location (or across the campus) for data protection. SCSI is “native”, meaning that systems can easily boot from SAS subsystems (including RAID-based booting for critical servers and dual-port booting for failover), whereas booting from an iSCSI subsystem is not always simple or even possible.
SAS uses relatively inexpensive wire for connection. FC can, in theory, use wire, but in practice must use optical fiber, and is required to do so in order to obtain its highest speeds. iSCSI typically uses wire (your Ethernet LAN cabling), although it might typically run over optical fiber to obtain its top speed – but those 10-gigabit/second iSCSI network components are priced up there with the 4GB/sec FC components.
I’m about to break a promise here and get into feeds and speeds. SAS runs at 3GB/sec. That seems slower, but SAS ports can be ganged together as trunks to yield higher speeds. OK, enough of the speed/feed specs. What counts is real-world performance. Examples of SAS performance equal to that of FC in mission-critical large-scale applications (in excruciating detail) are available on the Storage Networking Industry Association (SNIA) Web site, www.snia.org. Have your CIO check them out.
High-performance and reliability at low cost using commodity parts and full program compatibility is the natural trend of design “evolution.” (Although biological evolution involves no design or designer, the simile works because we care about results – in both nature and design – the fittest prospers.) SAS is the future of disk subsystems, with great benefit to you, the business end-user.
Dan Tanner is an independent analyst and consultant, and founder and principal of ProgresSmart (www.progressmart.com). He is an Individual Contributor to the Storage Networking Industry Association (SNIA), President of the New England Chapter of the Association of Storage Networking Professionals (ASNP), and a member of the Storage Network User Group of New England (SNUG/NE). [Note: The ASNP was merged into SNUG in May 2007.] From April 2003 to December 2003, he was Vice President, Business Development, KOM NETWORKS and lead in bringing technology and market potential awareness of the company’s products to the marketplace and business partners. From November 1999 to April 2003, he was Director of Storage Research, Aberdeen Group, Boston MA.
