Author: Susan Bobholz, Intel Storage Group Ecosystem Enablement Manager
and member of the STA Board of Directors
The first generation of Serial Attached SCSI (SAS) products represents a major leap forward in storage technology. SAS drives have a full-duplex throughput of 3.0 Gb/s and are capable of providing a competitive advantage to enterprises that require fast access to data. While today’s servers also have 3.0 Gb/s bandwidth, servers soon will be capable of communicating with storage at 6.0 Gb/s. On the other hand, the availability of 6.0 Gb/s SAS devices is expected to lag behind that of 6.0 Gb/s servers. In order to bridge the anticipated gap, the storage industry is looking for ways to take advantage of faster servers by aggregating bandwidth to communicate with multiple storage drives, e.g., enabling a 6.0 Gb/s server to communicate with two 3.0 Gb/s or four 1.5 Gb/s storage drives simultaneously. Bandwidth aggregation would enable businesses to take advantage of faster servers while still protecting their investments in current storage technology. Delivering this capability however, would have both technical and economic implications that must be considered.
Different approaches to bandwidth aggregation in SAS are being discussed and considered. One proposed approach is connection multiplexing, which would enable a server to split its communications across multiple logical links. As an example, a SAS initiator could negotiate with an expander to establish the multiplexed protocol, and a 6.0 Gb/s HBA or RAID controller could then communicate with two (3.0 Gb/s) SAS drives. This is an attractive solution because it would enable current storage technology to take advantage of higher bandwidth servers and operate more efficiently.
However, connection multiplexing would also have some significant draw-backs. Implementing the concept of logical PHY would require a change to the PHYs (physical layers) of both SAS initiators and expanders, resulting in intrusive alteration of all future SAS initiator and expander ports. The economic as well as technical costs of such changes would be quite high. Economically, two independent components, SAS initiators and expanders, would bear costs to make relatively complex changes. In addition, a fundamental change in SAS ports would force businesses to upgrade all their equipment at once because backward compatibility would be lost.
Technologically, there would be impacts on the evolution of storage eco-system components. Ideally, components would be able to evolve independently and not hinder one another’s adoption. But with two separate components requiring changes, they could no longer operate independently. Complexity would be added because data buffers would be needed on each port of both components to manage the connection, multiplying as the number of ports grew. The higher economic cost could slow the adoption of SAS and threaten its adoption on the motherboard.
Because of these high economic and technological impacts, connection multiplexing may have too high a price tag.
A different approach, buffered expanders, would provide a better solution. A buffered expander is similar in concept to connection multiplexing in that it enables bandwidth aggregation across multiple drives, but it would not require changes to both SAS initiators and expanders. Instead, link aggregation would be handled exclusively by the expander. The expander would manage the buffering of connections between SAS initiators and targets and other components would remain unaffected. The major advantage of this approach would be that only one eco-system component would be affected, and future SAS elements could continue to evolve independently. By limiting the impact to the expander, the overall cost of adoption for consumers would be minimized because it would preserve backward compatibility. Businesses could preserve their current SAS investments so the future adoption of SAS technology would not be threatened.
While this solution has several benefits, it is not without cost. The cost and complexity of expanders would increase. However, only those expanders that provide the buffering capability would be affected. Current expander products would not be impacted and would provide the same lower cost solution they do today. With the availability of different types of expanders, businesses could choose the product that is right for their specific environment.
The Right Path to Bandwidth Aggregation
Bandwidth aggregation would enable SAS devices to take advantage of 6.0 Gb/s server speed even though 6.0 Gb/s SAS drives are years away. But smart evolution decisions concerning bandwidth aggregation must be made carefully. Bandwidth aggregation would be a logical approach to enabling efficient communications between servers and storage devices that operate at different bandwidths, but the application of this technology should have a minimal impact on both the end-user and the future evolution of technology. Buffered expanders offer the best approach in terms of economic and technology impact and would support the independent evolution of future storage technology components.