Author: Linus Wong, Director Strategic Marketing and Tom Treadway, CTO
Adaptec
For the past five years, your only choice for developing a high-availability, highly scalable and reliable storage subsystem has been to use a Fibre Channel (FC) storage area network (SAN)-based architecture. A number of key features have established FC as the platform of choice for the highest storage performance. It provides:
- Enterprise-class reliability
- 1 and 2Gb/s bandwidth, with 4Gb/s products entering the market now
- Support for multiple topologies
- Scalability up to hundreds of servers
- Flow control to eliminate congestion
- Mature set of storage network services
However, FC has not proven to be a universal solution because its price point and maintenance requirements have put its level of reliability out of reach of smaller organizations. For businesses that need low-end networked storage, Serial Attached SCSI (SAS) can provide similar reliability at a far more affordable price.
Although SAS was originally designed to replace the aging parallel SCSI interface as an inside-the-box connection in direct-attached storage (DAS) implementations, it actually has many of the qualities of a network fabric. For example, it scales easily without downtime by using expanders the way switches are employed in FC architecture. Like all multi-node environments, SAS allows multiple hosts to access various storage systems and also provides redundancy, offering multiple paths for a host to access its data.
The table below shows additional points of comparison between FC and SAS as a network fabric.
| Fabric Attribute | Significance | FC | SAS |
|---|---|---|---|
| Long cables | In a data center, the computers and drives may not necessarily be stacked right on top of each other. Cables don’t have to be long enough to go across campus, but they must at least be several meters long. | Yes, up to 30m copper and 10km fiber optics. | Yes, up to 8m |
| Fault-tolerant cabling | A bus cable, used by parallel SCSI, is impractical because there are multiple devices on a single set of wires. A point-to-point architecture is more forgiving of cable failures. | Point-to-point | Point-to-point |
| Flexible cable configuration | Beyond point-to-point, the fabric should be able to support topologies more complex than just a loop with each device having an “in” and an “out”. Interconnected tree topologies are extremely versatile. | Hierarchical, i.e., tree | Hierarchical, i.e., tree |
| External cables | Cables must be well-shielded to maintain signal integrity outside of the computer chassis | Yes | Yes |
| Wide pipes | Any good topology has a mixture of pipe sizes to accommodate varying bandwidth requirements | Not in practice, though the 1Gb, 2Gb and 4Gb speeds can be considered as different pipe speeds | Yes, wide links |
| High quantity of devices | The topology should support a multitude of drives | Practically unlimited, using fabric addressing | 16,384 in one SAS domain, with unlimited domains |
| SCSI-command based | This refers to the command set, not the physical protocol | Yes | Yes |
| Efficient topology discovery | During boot, the hosts and drives must quickly find the multiple paths to each other. The devices must identify illegal topologies, such as loop-backs | No, fabric addresses typically entered via management software | Yes |
| Topology change notification | Topology changes, such as a device or switch being inserted or removed, must be communicated to the host and drives | Yes | Yes |
| Cost options | Should provide a wide range of solutions, from expensive/fast to cheap/slow, with many different port counts, etc. | Numerous solutions, but rarely inexpensive | Yes |
| Interoperability | With so many potential devices in the SAN, it is unlikely that they will be sourced from a single vendor, so the ability to work together is key | Not in the past; sufficient today | Yes |
One example of a storage interconnect scenario that can benefit significantly from the use of SAS is creation of shared storage in a blade environment. At a time when the use of blade servers is expected to grow 56% annually*, SAS provides an ideal way to connect blades with a shared storage pool.
In the SAS scenario, expanders provide low-cost interconnection between the blades and the shared storage. SAS also provides the advantage of a SCSI DAS boot model in which each blade server can see a dedicated boot drive, even though those drives are physically located in a separate, storage “brick.”
As mainstream storage has increased in complexity, it has outgrown the two-node cluster practical limit of parallel SCSI. Today’s mainstream storage applications demand high scalability, high reliability, and ease of management, yet the FC SANs that provide these qualities remain prohibitively expensive for most of the market.
SAS offers a cost-effective alternative that is ideal for delivering high-end clustered solutions to the mainstream market. It provides benefits similar to a FC SAN at a much lower price point. It also allows the user to more finely tune cost and performance parameters to storage needs by integrating either low-cost, high-capacity SATA disk drives or high-performance, high-reliability SAS disk drives.
*IDC Worldwide and US Blade Server 2005-2009 Forecast and 2004 Vendor Shares; published July 2005
