- Avoid overspending and overprovisioning.
It's obvious that spinning disk is the source of most storage power consumption, and unused spinning disk represents wasted energy. But there are several challenges in growing storage incrementally, including the organization's ability to accurately forecast storage capacity needs. Beyond capacity planning, it requires a relationship with a vendor who can support the incremental storage growth. The technology must also allow the storage to expand easily and with minimal disruption.
- Revisit your tiering strategy.
Small, fast disks demand more energy than large, slow disks, so the distribution of data across various storage tiers can have a big impact on your power consumption. A study by EMC Corp. last year indicated that storing a terabyte of data on a 7,200 rpm 1 TB SATA drive is 94% more efficient than storing it on a 15,000 rpm 73 GB Fibre Channel (FC) drive. This provides an added incentive to ensure that storage service levels and tiers are properly aligned based on application and data value.
There are many situations where tiered storage allocations are far from ideally distributed. Understanding this distribution and developing a clearly defined set of service-level requirements to apply to new and existing applications can lead to substantial savings in equipment cost and energy use.
- Revisit RAID policies.
Another facet of a tiered storage strategy is the RAID data protection policy applied to a given tier of storage. It's not so prevalent these days, but overprovisioning of RAID 1 or RAID 10 increases the number of spindles and power consumption. When additional performance or availability isn't required, the drive count can be reduced.
- Consolidate storage arrays.
The more frames in a data center, the more power that needs to be reserved for them. Older arrays tend to be less efficient than the latest models, so reducing the number of storage systems is an obvious option, just as consolidating physical servers through virtualization is. Newer virtualization arrays may also offer enhanced functionality such as thin provisioning to improve utilization and further reduce energy consumption.
- Establish power usage metrics.
Power specifications typically enter a storage infrastructure discussion only during installation planning. However, to truly align data center and IT infrastructure objectives, this will need to occur. Organizations like The Green Grid promote standardized data center efficiency metrics. It will also become necessary to establish metrics such as GB/kW or IOPS/kW, and to then determine these rates for each tier of storage as well as usage for the storage infrastructure in total.
- Consider solid-state drives (SSDs).
While solid-state drives (SSDs) are still a new phenomenon for many, they can play a role in replacing power-hungry, high-speed, low-capacity disks. In addition to offering higher performance, EMC reports that on a per-IOPS basis, flash disks require 98% less energy. It's therefore imperative that organizations do their homework and understand actual performance requirements before making any substantial investment in this technology.
- Consider massive array of idle disks (MAID).
At the other end of the performance spectrum is MAID technology. We know a significant amount of data currently stored on spinning disk is accessed infrequently. From an energy-consumption standpoint, the most efficient disks are the ones that aren't spinning at all, which is the rationale behind MAID. For archival data that still requires accessibility, this technology represents an attractive alternative to conventional, continuously spinning, nearline storage.
- Make energy usage a buying consideration.
As the demand for green storage grows, vendors have realized that energy efficiency can be a competitive differentiator. It makes sense to factor this into equipment purchasing criteria, and to consider energy impact when architecting new storage infrastructures. Often times, the focus is on capital expenditures (CAPEX) and insufficient attention is paid to operation expenditures (OPEX). However, evidence is mounting in the server world that these lifecycle costs, including power and cooling, can actually overshadow CAPEX.
This was first published in June 2009