Author: Ahmad Zamer
On a recent trip to the North East, I had the opportunity to visit the data centers of two corporate customers. The first was an Internet Service Provider (ISP), the second, a mid-size brokerage firm. Both organizations, despite their different businesses, spoke of similar challenges their Information Technology (IT) staffs face. They also spoke of similar needs they wanted to be able to meet using new technologies.
Real-World Data Center
The ISP’s manager of data center operations spoke in terms that may not be unfamiliar to technical people who work on industry standard and storage product design. He talked about his needs and challenges in terms of how they relate to floor tiles. He measured the value and utility of every storage and server solution in terms of how much precious real estate they would occupy in his data center. Following a three-hour tour and lecture from the data center director, one finds it hard not to measure everything using the “tile factor.”
From the ISP’s perspective, the major challenges were the limited power, per tile, they can bring to the data center, and the ever-rising costs of cooling per tile. These are simple metrics that have serious implications for all new computing technologies, especially storage technologies. Pointing to huge boxes scattered across the data center, our guide shouted, “Do you see these huge boxes? These are my storage boxes and I don’t seem to ever have enough.” So, how can one keep adding big storage boxes to the enterprise infrastructure without negatively impacting the tile factor?
Power and Heat Trade-offs
Serial Attached SCSI (SAS) offers a solution for the above dilemma. Mindful of higher hard-drive power consumption and heat dissipation, SAS supports a smaller hard disk form factor of 2.5 inches. The new small form factor (SFF) drives consume about 40% less power, compared to the larger 3.5-inch drives, and dissipate much less heat. These features enable data centers to increase the density of storage per tile without bursting the envelope of power consumption or breaking the bank with cooling expense. SFF drives also allow IT staff to increase the number of spindles in their racks while preserving storage capacity. SFF does not mean smaller capacity. New SFF drives will soon become commonplace in the data center with drive capacities comparable to those of 3.5-inch drives.
At the brokerage firm, the story was similar. Power consumption and scalability were important, but performance was the main issue. The company IT staff was most concerned with the response time of their systems at peak usage times. It was essential to avoid any degradation of response time when client requests peaked at unpredictable times. The staff was concerned that adding larger pools of storage results in lowered performance at peak times. Again, this is a seemingly simple need that also has serious implications to how we design new storage technologies.
Performance of storage systems is addressed in SAS to ensure that customers continue to get the response time and throughput they need for their applications. SAS is a point-to-point architecture that enables discrete and separate data paths for each drive connected to the controller. Unlike shared bus environments found in legacy SCSI, every disk connected to SAS has the full 3Gb/s data rate of the SAS link. In fact, thanks to the full duplex nature of SAS and dual-ported hard drives, the point-to-point topology makes a 6Gb/s data pipe available for every drive. That effectively doubles the data transfer rate of the SAS links.
Addressing the reality of today’s storage needs, SAS increases the number of devices that can be connected to 128 (per edge expander) from the 15 supported by the parallel SCSI bus. When a fan-out expander is used to aggregate multiple edge expanders, a single SAS domain can contain more than 16,000 drives. Expanders are discussed in more detail elsewhere in this issue of Serial Storage Wire.
Since high availability tops the list of “must haves” for all enterprise data centers, SAS ensures that hard drives are hot pluggable. With that feature, IT staff can expand storage or replace failed drives without impacting the operation of their data environment. Hot swapping allows users to expand, alter, or service their installations on the fly. Data centers can scale their storage to meet their needs using SAS.
In addition, SAS provides for wide ports. In order to create a wide port, two or more physical levels (PHYs) are combined to create a SAS port that delivers the aggregate bandwidth of all the PHYs. It is typical to have four PHYs in one wide port, enabling an unprecedented 12Gb/s per port. With that potential bandwidth, customers are assured of having the best response time possible from a storage solution.
The serial SAS interface is more efficient than the parallel legacy interface that it will replace. A serial interface is not encumbered by data skew, shared bandwidth and termination anomalies. As a result, users can expect improved performance just for switching to Serial Attached SCSI.
To further enhance the high performance of SAS drives, the protocol adds a suite of advanced data-handling features. Command queuing is a SAS feature that offers 256 queue levels with intelligent data handling properties that speed up data transfers and facilitate data flow. As one might expect, SAS utilizes top-of-the-line 10KRPM or 15KRPM hard drives, the latter with the lowest access times of any hard drives. That places SAS disk performance on a par with the best available storage technologies.