Sorting Out the Options
Since then, hard drives have become a little more compact, faster and dense and there are certainly more options. The most common drives found in todays data centers are parallel SCSI (small computer system interface), ATA (advanced technology attachment) and FC (Fibre Channel).
The latest drive technology to arrive is SAS (serial attached SCSI).
SCSI drives have been around since 1986 when the SCSI protocol was standardized. Over the years they have come to dominate the server-class storage market. As with most technology, they have become smaller in size, larger in capacity and faster.
Todays largest Ultra320 SCSI drive (probably the last of the parallel drives to be developed) is 146GB in capacity, 15,000 RPM, and has a bandwidth of 320MB/sec.
The biggest limitations with parallel SCSI is the number of drives supported (15 per bus) and the distance parallel SCSI can go (12 meters). These limitations cripple the effectiveness of parallel SCSI in todays storage-dominated datacenters but they work well as internal drives for servers. It is expected that parallel SCSI drives will all but disappear from the enterprise by 2008.
Integrated dive electronics ATA drives (IDE ATA), also know as parallel ATA (PATA), can be found in almost every PC ever made. Early adopters of intermixing ATA drives with FC drives in a single storage system used PATA drives.
These were quickly replaced with the newer, faster and larger SATA (serial attached ATA) drives. SATA drives are typically used within an enterprise to provide large (multiple terabyte) pools of storage to hold secondary, tertiary and n-ary copies of data.
These copies are usually used for backup and archival of data. SATA drives are not intended to replace FC or parallel SCSI drives; desktop-class SATA drives have a mean time between failure (MTBF) of about 600,000 hours with a duty cycle of 8x5 (8 hours a day, 5 days a week), FC drives have a MTBF of approximately 1.2 million hours with a duty cycle of 24x7.
So how does that translate to the real world? It means that a FC drive can stand up to the rigors of a large database doing a large amount of transactions against it for much longer than a SATA drive would; and as a result, SATA drives should not be used in this environment.
SATA drives are also much slower than SCSI and FC drives and are not a good fit for high-access, high-performance data.
Fibre Channel drives have only been around for about six years and compare to parallel SCSI drives in performance, reliability and capacity.
The biggest difference between parallel SCSI and FC drives is how many drives can be supported in a bus or channel. Parallel SCSI supports 15 drives per bus while FC supports 127 drives in a loop or about 16 million in a fabric.
This scalability allows vendors to easily architect and deliver large-scale, high-availability storage devices without the complexity of multiple SCSI busses. For example, an entry-level, redundant FC array with 112 drives needs just one pair of redundant FC loops for the drives; a similar SCSI array would need at least eight busses.
SAS drives first came to market in 2004 and is the replacement technology for parallel SCSI. SAS drives, like FC drives, are equal to parallel SCSI drives in performance, reliability and capacity. SAS doubles the distance limit of parallel SCSI and increases the number of drives on a bus from 15 to over 120.
SAS also has a short-term advantage over other drive technologies in that 2.5 inch drives are available now; FC, SATA and parallel SCSI drives are 3.5 inches. This means that SAS can deliver more capacity in a given space than FC or parallel SCSI.
Now that drives are abundantly accessible, more SAS solutions are available. There are three ways that SAS can be deployed in the datacenter: direct attached, in a small SAS storage area network (SAN) or as part of a FC SAN.
Direct Attached Storage
Over the past few years, trade magazines and storage vendors have been loudly touting SAN, network-attached storage (NAS) and more recently, iSCSI. Meanwhile, the tried and true direct attached storage (DAS) solutions are ignored.
DAS, however, is an excellent solution for a couple of servers that need to have their own storage or a shared pool of storage. DAS-based SAS solutions offer excellent cost-effective solutions to FC and parallel SCSI alternatives. The main advantage to FC is cost; the advantages over parallel SCSI are performance, scalability and investment protection.
SAS-based Storage Network
A SAS-based storage network is a storage network that uses SAS as an interconnect instead of FC. There are now SAS switches on the market that support connectivity between multiple hosts and storage devices.
While FC SANs easily support dozens of storage devices, servers, tape libraries and appliances (NAS, iSCSI, VTL, etc.), SAS storage networks are only intended to support about 12 servers within a small footprint (defined by SAS distance limits); a perfect size for many companies needs.
Fibre Channel SAN
SAS deployed as a DAS solution is an excellent solution for small organizations or point solutions (i.e., a dedicated disk pool for a disk-to-disk backup solution).
Deployed as a SAS-based storage network, SAS is a good fit for some small and medium enterprises, but SAS also fits well into a large enterprise environment. Some FC arrays support SAS disks. This allows the high bandwidth, low latency, large capacity SAS disk to be deployed anywhere within the enterprise and integrate seamlessly with a new or existing SAN.
FC attached SAS is not intended to replace FC or SATA disk, it is meant to enhance existing solutions and offer more options when designing solutions.
Parallel SCSI is all but dead; FC is an excellent solution for high-end, reliable storage needs; SATA works well for backups and archive. For everything else, there is SAS.
Whether theres a single server, a small group of servers or a large enterprise-class SAN, SAS is a viable solution. Because of this, analysts predict that SAS storage market share grow steadily through 2008, replacing the parallel SCSI market share in the enterprise storage market, while FC and SATA market share will remain flat.
Jim McKinstry is senior systems engineer with the Engenio Storage Group of LSI Logic, an OEM of storage solutions for IBM, TeraData, Sun/StorageTek, SGI and others.