Operating Systems

Mar 8 ’10

Storage system design always involves a trade-off between cost and performance. A storage system full of small, fast disks may provide great performance and throughput, but the benefits don’t always justify the costs. “Tiering,” or using a mixture of faster and slower disks, is likely to produce a better price-to-performance ratio. You can reserve the fastest, most costly disks for the workloads that really need them and let large disks handle less active workloads. 

Tiered configurations are more challenging to design and manage, but with care, they can be significantly less expensive and offer great performance. This article examines how to assess whether tiering can be used in your environment to reduce storage acquisition costs while maintaining the required performance levels. It also explains how to design a tiered solution with confidence. 

Potential

When carefully designed, multi-tier solutions can be much more economical. The savings will be greater when there’s a mix of highly active workloads with high-performance requirements and less active workloads with less stringent performance requirements. In such situations, a multi-tier storage system may yield a less costly and faster solution than is achievable with a single tier. 

Most of today’s mainframe installations predominantly use 146GB 15k RPM Fibre Channel (FC) disks. When acquiring new hardware, there are many more options available, including playing it safe with the same drive technology or moving everything to larger drives. If you can altogether move to drives that are twice as large, only half the number of drives will be needed, significantly reducing acquisition, maintenance, and environmental costs. Alternatively, if one-third of your data is quite active and needs the 146GB drives, it may be possible to store the remaining data on disks as large as 450GB. This combination of disks also would decrease the number of Redundant Arrays of Independent Disks (RAID) array groups to a little more than half the number of original groups. An original configuration with 36 146GB RAID array groups would have just 12 + 24/3 = 20 RAID array groups in such a new configuration. 

Of course, one RAID array group of larger capacity disks is often more expensive than one RAID array group of smaller disks, so halving the number of RAID array groups may not halve the acquisition costs. As prices between larger and smaller disks vary per vendor, we’ve estimated the relative costs listed in Figure 1 for the examples in this article. Based on the numbers in the relative costs table, the cost difference between the tiered solution with 12 groups of 146GB disks and eight groups of 450GB disks, and the original setup of 36 groups of 146GB disks is 28 compared to 36, a 22 percent reduction in acquisition cost while still using 15k RPM technology. For a more detailed example of the possible cost savings, see the tiering examples in the accompanying sidebar.

Risks and Drawbacks 

The drawback of a storage configuration with multiple tiers is that it’s more complex to manage. Once you manage multiple tiers, you need to decide which data needs to be handled by which drive technology. These decisions are crucial for the effective performance of the multi-tier configuration; mistakes can compromise an otherwise good design.

In smaller configurations, the separation of data into tiers makes it more challenging to effectively use drive capacity. If you can’t effectively use the space, you might need to configure more RAID arrays, precluding some of the cost savings associated with multi-tier. 

5 Pages