Save Money and the Earth by Lowering Data Center Power Consumption

By Tom Clark, SNIA Board Member, Principal Engineer, Brocade

Among all the challenges CIOs and IT administrators currently face, two historical trends are on a collision course that threatens widespread collateral damage. On the one hand, the geometric growth in data processing is generating ever increasing demand for servers, storage and the infrastructure to support them. More data means more hardware, larger data centers and the requisite increase in power and cooling to sustain continuous operation. On the other hand, the limited availability and increasing cost of energy worldwide is undermining the energy utilities' ability to supply reliable power. Competition between developing and developed countries for limited energy resources, regulations to reduce environmentally harmful emissions from fossil fuels such as coal and oil, and lack of accelerated investment in sustainable energy sources are starkly pointing to an impending conflict between projected supply and demand. Given that modern enterprises and institutions all depend on information technology for their livelihoods, finding the means to align energy consumption to energy availability and simultaneously accommodate data growth are now critical components of a viable IT strategy. With today's technology you can build a sophisticated data center fairly cost-effectively but even a generous budget will not guarantee that you can plug it in. According to a recent Gartner forecast, half of the data centers worldwide will not have sufficient power and cooling by the end of 2008. ¹

In response to this approaching data center crisis, IT industry vendors (the majority of which are Storage Networking Industry Association [SNIA] members) are participating in a new Green Grid initiative in recognition that "...energy efficiency in the data center is the most significant issue facing technology providers and their customers today."² This statement represents a radical shift in IT focus that traditionally has concentrated on optimizing compute resources and storage operations. The Green Grid consortium takes a holistic view of the entire data center ecosystem and is thus examining all elements that contribute to energy consumption and waste. Systems that generate excessive heat, for example, require air conditioning systems that in turn consume more power. The heat dissipation represents inefficient use of power by the system; the cooling represents an inefficient use of power to offset the original inefficient use of power. To avoid compounding the problem, new means must be found to resolve core inefficiencies and thus relieve the overall power burden.

Although The Green Grid's charter includes all IT data center assets, a specific focus on green initiatives for storage and storage networking is being developed by the SNIA. The SNIA Green Storage working group is concentrating on the carbon impact of the complex of servers, switches, directors, storage arrays and tape subsystems that compose data center storage networks. As the umbrella organization for storage-related hardware and software vendors, resellers, integrators and end users, the SNIA will focus on raising consciousness around environmentally responsible storage networking practices and identify means to resolve the contradiction between data growth and growing power scarcity.

Going green in the data center has many facets, including reduction of overall power consumption, more efficient utilization of the power used, reduction of hardware via consolidation, aligning storage to data requirements and decreasing the total amount of storage required to meet data needs. As this multi-faceted approach implies, there is no single solution for optimizing data center energy efficiency. Collectively, vendors and customers will need to implement comprehensive strategies that cooperatively integrate hardware, software and operational elements.

Compared to consumer products that may more readily lend themselves to energy-saving features (e.g. EPA Energy Star compliance), a data center infrastructure poses considerable challenges in achieving energy efficiency. Business today is global and data centers must typically support 7x24x365 operations. Data transactions demand immediate response times to meet both business and customer satisfaction requirements. Because continuous operation is essential for the viability of an enterprise, there are few idle elements in a storage network that could leverage, for example, low-power hibernation techniques typical of consumer electronics.

Data centers are inherently power-hungry. The challenge, then, is to temper the energy appetite of data center operations and maximize data transaction output per unit of energy consumed. Some advances have already been made on this front, as demonstrated by vendors who have proactively incorporated power efficiency into their product designs. In addition, new storage technologies such as server and storage virtualization, information lifecycle management, storage compression and data de-duplication are enabling more efficient use of storage assets and overall reduction of the carbon footprint.

Given the sheer volume of file and application servers used by today's enterprises, servers are a prime candidate for a green makeover. More efficient AC to DC conversion, reduced heat dissipation and more efficient use of CPU cycles by individual servers have a significant impact when multiplied by the thousands of servers typical of large data centers. By enabling multiple instances of operating systems and applications to be hosted on a single server hardware platform, server virtualization promises to reduce the total amount of hardware and associated power consumption to service business applications. Blade server architecture likewise significantly reduces the hardware footprint required to support applications, and depending on the supplying vendor may have greater energy efficiency than rack-mount or stand-alone servers. The combination of concentrated blade servers and server virtualization software can thus support more data transactions on less energy consuming real estate and help reduce the circumference of the data center and its cooling requirements.

For the storage network infrastructure that ties servers to storage, the complex of switches and directors typical of large data centers represents another green challenge. Data sheet power ratings may vary from actual usage depending on how the storage network is designed and what layers of connectivity must be supported. The current trend in data center consolidation is to eliminate more fabric elements by collapsing connectivity into larger port-count fabric directors in a tiered, core-edge design. In addition, auxiliary storage services for multi-protocol support, distance connectivity, high speed inter-switch links, fabric-based storage virtualization and data migration facilities are being integrated on more sophisticated director platforms. The larger chassis required to support port connectivity and advanced storage services should therefore be designed for optimum energy efficiency, particularly for enterprise data centers that deploy hundreds of directors supporting thousands of servers and storage devices.

For storage arrays, a number of options exist to minimize energy usage. Individual disks drives may be relatively cheap, but each spinning disk represents continuous power consumption. Typically, the faster the disk, the more power consumed. Mission-critical applications with high availability and high performance requirements may indeed need the very fastest disk drives and full mirroring (doubling the total number of disks) to sustain operations. The data generated by those applications, however, may age with declining business value. By combining information lifecycle management to track the value of data at any point in time with tiered classes of storage it is possible to migrate data from a higher energy usage asset to a lower one. A second tier array using lower speed SATA drives, for example, would be a more energy efficient repository for lower value data before the data finally retires to an even more efficient media such as tape. In addition, new technologies such as MAID (massive array of idle disks) provide on-demand access to disk data with the majority of drives in an array spun down until a data request is received.

Other green storage options attempt to reduce the total amount of storage required to house corporate data. While data growth has been a constant for all enterprises and institutions, a significant portion is due to redundant copies of data dispersed throughout the network. For file-oriented applications, in particular, identifying and eliminating redundant copies of files can dramatically reduce total storage requirements. Implementing a global name space, for example, can facilitate elimination of dispersed file silos and contain the tendency to replicate data as a convenience for local access. Likewise, data compression to disk and data de-duplication technologies can preserve data accessibility while reducing the amount of storage, disk drives and accompanying power usage required.

Data redundancy is also an inherent inefficiency for companies with geographically dispersed sites and remote offices. If every remote location has its own storage, green storage issues are multiplied across the corporate network. The current trend in remote storage consolidation is to centralize storage assets at an optimized central data center and leverage WAN acceleration to service file access.

The former metric of cost per gigabyte of storage capacity was limited to hardware and operational costs. The green metric is kilowatt cost per gigabyte, or more appropriately, kilowatt cost per disk. Even spare drives in a typical storage cabinet are always on, always spinning, always drawing power and always dissipating heat. Implementing tiered storage, storage virtualization, lifecycle management, data compression and de-duplication can help lower the carbon impact of a data center and reduce overall costs.

In response to the rising cost of energy and its limited availability, some large enterprises are simply relocating to areas with lower energy costs. New data centers being built along the Columbia River in Washington, for example, are taking advantage of the relatively lower cost of hydroelectric power compared to conventional coal or oil-based power generation. While this does save money and lessens the environmental impact vis-à-vis carbon-emitting power sources, there are still environmental issues concerning hydroelectric power and the storage infrastructure within the data center. Going green means that all aspects of IT operations including facilities, people and infrastructure must be reexamined to identify areas where greater power efficiencies can be achieved and around which proactive best practices can be established.

The SNIA encourages all storage networking vendors, channels, technologists and end users to actively participate in the green storage initiative and help discover additional ways to minimize the impact of IT storage operations on power consumption. If, as Gartner forecasts, adequate power for many data centers will simply not be available, we all have a vital interest in reducing our collective power requirements and make our technology do far more with far less environmental impact.

¹ "Gartner Says 50 Percent of Data Centers Will Have Insufficient Power and Cooling Capacity by 2008," Gartner Inc. Press Release, November 29, 2006.
² "The Green Grid Opportunity" White Paper The Green Grid 2007

By Robin Harris, Data Mobility Group and StorageMojo.com

As scary as lions, tigers and bears? Maybe blocks should be.

Why do we manage blocks? You might say we manage blocks because disks have blocks and we build storage out of disks. But what if disks didn't have blocks? No more block management. We'd simply manage ... OK, what would we manage?

By Dr. Kevin McIsaac, Advisor, Intelligent Business Research Services (IBRS)

Introduction
With storage requirements typically growing at 35%+, most organizations are finding they must routinely add more capacity. To avoid creating silos of storage capacity that can not be shared and optimized, or building a storage infrastructure that becomes increasing complex and costly to manage, IT organizations must properly plan and execute storage acquisition.

Observations
Often, new storage acquisitions result in a sub-optimal storage infrastructure as the requirements are driven by the needs and demands of a new high profile application implementation, e.g., ERP, e-commerce, e-mail archive. This often results in a storage infrastructure that can not be shared, or optimized, across the application portfolio because the planning does not take into account the needs of other applications, either existing or planned.

With networked storage infrastructure now mature, and rapidly commoditizing, organizations can gain many benefits from creating a shared storage infrastructure, e.g., lower complexity, shared processes, higher staff efficiency, more effective disaster recovery.

Phase I - Requirements Analysis
The first step, and the key to getting storage acquisition right, is the requirements analysis. A simple and highly effective approach is to create a portfolio of the major applications and for each application capture the most important high-level storage attributes for the data associated with that application. This list should include both existing applications and any new applications that are planned for the next 3-5 years.

The analysis should only take a few days, providing just enough information about the data to enable a broad understanding of the storage service level requirements. A more detailed analysis of the data, such as the syntax or semantics, is not necessary to understand the storage system needs.

Rather than using typical technical metrics (e.g., spindle speeds, latency, cache size, attachment method) it is more valuable to focus on high-level, business-oriented "Storage Service Levels" such as Hours of Operations, Recovery Time Objective (RTO), Recovery Point Objective (RPO) etc., along with a few high-level technical metrics, such as capacity, growth rates, data type and the application "use case."

The typical storage service metrics collected during the requirements analysis phase follow:

• Use case - this is defined by how the application uses the data. Common use cases are :
  Transactional - data with high volume of random reads and writes. Examples include applications based on databases, such as Oracle, MS SQL Server, or MS Exchange.
  Read mostly - data that is changed infrequently but read frequently. Examples include files produced by office productivity applications, such as MS Word, MS Excel, etc.
  Reference - data that is written once and then referenced by many applications over a long period. These data may need protection from change and a retention policy.
  Archival - large sets of data that do not change and are read infrequently. Over time these are less likely to be accessed and are not needed as quickly. Archival data may need protection from change and a retention policy.
• Size - the usable storage capacity required by the application, measured in GB or TB.
• Growth - the compound annual growth rate of the data measured as a percentage.
• Recoverability - typical metrics are:
  Recovery Point Objective (RPO) - the point in time prior to the outage that the data is restored to, e.g., 1 hr, 1 day, 1 week.
  Recovery Time Objective (RTO) - the time required to restore access to data after an outage, e.g., 1 hr, 1 day, 1 week.
• Availability - Typical metrics are
  House of operations (HOO) - typically measured as 7x5, 10x6, 24x7 etc.
  Mean time to failure (MTTF) - e.g. 1 year, 3 years.
• Retention - the period of time that data is to be retained.
• Performance - an indication of the performance required by the application. This should be characterized as online, nearline, or offline modes, with a sub characterization of fast, medium or slow. This can be supplemented with an indicative time, e.g., nearline - fast (< 1min). For near-line or off-line data, this is a measure of the time to first byte.

The service levels should be defined by the business based on the value of the application to the business operations. Often this has not been formally defined before, and typically the business expects unrealistically high levels of service. At this stage, the IT organization should help the business understand what levels are realistic, and then re-examine these choices at the final stage once costs are better understood (see "Service Level Negotiation" below).

The storage service levels for the application portfolio are collected in a spreadsheet (Figure 2) to create a Storage Requirements List. Applications with similar storage requirements are grouped together defining "Storage Tiers," e.g., Bronze, Silver, Gold etc. These storage tiers should be formally defined in a simple table (Figure 1).

Figure 1 - Storage Tiers

Figure 2 - Storage Requirements List

Finally, extend the Storage Requirements List with 3-5 years of application capacity estimates, and create a summary Storage Requirements Plan that is broken down by Storage Tiers (Figure 3).

Figure 3 - Five Year Storage Requirement Plan

These three tables define the storage requirements, and provide sufficient information to define a high-level storage solution with cost estimates.

Phase II - Solution Mapping
In this phase, storage tiers are mapped to specific storage technologies and configurations based on the required storage service levels. This is done by understanding which technologies and configurations are able to meet the specified service levels, while optimizing cost or some other consideration.

For example, using the Storage Tiers in Figure 1:

• The Gold tier is mapped to a SAN attached mid-range storage array using high performance disk (15K FC-disk) using RAID 1. Disaster recovery is based on synchronously replicating the data to a secondary array at the recovery site.
• The Silver tier is mapped to the same SAN attached mid-range storage array using mid performance disk (10K FC-disk) with RAID 5. Disaster recovery is based on nightly tape backup, using a robotic tape library that is located at the disaster recovery site.
• The Bronze tier is mapped to a Network Attached Storage (NAS) head that leverages the mid-range storage array and uses low cost disk (i.e., 750GB SATA) with RAID 6. Disaster recovery is based on nightly, incremental tape backup, leveraging the robotic tape library that is located at the disaster recovery site.

Getting the solution mapping right, i.e., defining a technical solution that meets the service levels and optimizes cost, requires considerable expertise. When the requirements are simple or these skills are highly developed in-house, this can be done by IT staff. If not, it is highly recommended to engage an experienced storage consultant or VAR to assist with the solution design. Once this solution is defined, unit costs for each of the storage tiers can then be estimated, e.g., Platinum $30/GB, Gold $10/GB, Bronze $2.50/GB.

Phase III - Service Level Negotiation
Before finalizing the plan and purchasing storage, apply the estimated unit costs to the Storage Requirements List to create a provisional storage capital budget. Sophisticated IT organizations will use this with the business units to negotiate for trade-offs in the service levels for lower costs. For example, by renegotiating the service levels for e-mail from Gold to Silver with an e-mail archive policy that enables most e-mail data to be stored on bronze disk, the organization could save tens, if not hundreds, of thousands of dollars in capital expenditure.

The service level negotiation phase works best in organizations that have a culture of "user pays" chargeback.

Next Steps
Look for a trigger that will cause a significant storage purchase (e.g., new application implementation or upgrade, hardware end-of-life, change in application usage). Use this as the catalyst to kick-off the storage planning process.

Assess your current in-house storage skills and determine if external advice is necessary. If it is, carefully select a partner and involve them in the planning processes from the earliest stages.

About the Author
Dr. Kevin McIsaac, Advisor with the Australia-based Intelligent Business Research Services (IBRS - http://ibrs.com.au/), has over 20 years of IT experience and is a recognized expert in infrastructure, operations, vendor management and the art of running IT as a business. Previously, Dr. McIsaac worked for META Group, most recently as the Research Director Asia-Pacific, researching, distilling and disseminating best practices in IT. In that role he advised the CIOs and the IT management teams of leading Asia-Pacific organizations.

As part of the SNIA's ongoing commitment to hearing the voice of the end user, the organization's End User Council (EUC) conducts an annual survey, designed to prioritize the most important issues facing storage and storage networking end users today. The group recently released the full report of results from its third survey, titled Storage Interoperability: So What's the Problem?

Providing new levels of understanding of end user pain points, this most recent survey found several key insights from end users, including the fact that upgrades cause end users the most worry - more than integrating new hardware, interoperable management software, and outdated management software. The full report is available for download from the EUC Web site at http://www.snia.org/euc/.

Key Results
A significant finding of the survey is the extent to which the end user respondents desire interoperability. Figure 1 illustrates why this is important, depicting the number of vendors end users are supporting within each type of storage infrastructure in 2005 and 2006. With respondents' number of vendors increasing in most categories, this will continue to be a critical pain point for end users.

In fact, the survey found that 62% would pay more for an interoperable solution, and 23% would de-select a product that did not offer interoperability. Of the 62% who indicated a willingness to pay more for the interoperable solution, the average surcharge that was tolerable was 7.6%.

Another important finding came in answer to the question, "What are your biggest worries?" The top three responses to this question all related to upgrades, as shown in the ranking below:

1. Forced upgrades for legacy gear
2. Software upgrades
3. Firmware upgrades
4. Software storing data in formats unreadable by other applications
5. Management software for another vendor's hardware
6. New hardware integrating with existing SAN
7. Hardware not keeping up to date with operating system vendors
8. Management software not keeping up to date with hardware
9. Interoperability between same hardware OEM'ed with different vendors

The responses to the "biggest worry" question showed some divergence across the Large, Medium, and Small Business tiers. All companies ranked upgrades as high worry points, but the Small Business tier is more worried about firmware upgrades, while the Medium Business tier worries more about management software keeping up to date with hardware.

When asked to rate reasons for upgrading or replacing existing gear across the Small, Medium, and Large Business tiers, the need to "Increase Capacity" was the most commonly selected reason. There was a significant disparity, however, between the Large Business tier and the Small and Medium Business tiers on the second most commonly selected reason: "Rising Maintenance Costs." Large Business tier storage end users rank maintenance costs almost on par with increasing capacity, but in the smaller sized companies, there is a greater focus on getting more storage into the business as the key driver.

Other results include the role of testing, with just 17% of respondents indicating that they only trust what they test, and 29% indicating that they do in-house testing if time permits. Some 26% require purchase contracts to guarantee feature sets, and so presumably could call upon vendor resources to alleviate certification and testing costs because a failure would result in returned gear.

The relationship between end-users and vendors is very different across the Large, Medium, and Small Business tiers. Small Business tier respondents give much more credence to third party evaluations than Large or Medium Business tiers, which are unlikely to see these evaluations as more reliable than the vendor's assurances. While the Large Business tier is almost twice as likely to rely solely upon in-house testing, the Medium and Small Business tiers are almost three times as likely to say that they rely solely on a vendor's assurance as a means of saving time in testing and certification.

When asked to identify infrastructure tiers upon which they spent the most amount of time testing and certifying new equipment, 40% of the end user respondents indicated that "Storage Arrays" took the most amount of time to install, but then they ranked the array as requiring very little testing and certification time around maintenance upgrades. "Storage Management Software" was ranked by 30% of the end users as taking the most amount of time to install new.

In addition, when asked about the most important selections that guided a buying decision, "Vendor Support" and "Capabilities" were the two criteria garnering the most "Very Important" scores for a component - both within an existing SAN, as well as for a component in a new SAN. According to respondents, "Vendor Support" is an even more important criterion for a component in an existing SAN than for a new SAN infrastructure. The results also show that "Interoperability" and "Standards Compliant" is rated more importantly than the cost criteria of "Integration Costs", "Low Cost", and "Deployment Costs."

The results from this survey show an upward trend in the perceived barriers to new technology. Figure 2 shows the increase in each area, year to year.

While more users are feeling the impact of these barriers, it is significant to note that - as seen in Figure 1 - respondents indicate an increasing number of vendors across storage infrastructure tiers. Therefore, the "barriers" do not seem to have prevented an inflow of new vendors.

Conclusion
The SNIA End User Council (EUC) consists of customers interested in advancing the development of storage networking technology and solutions, and the group's annual survey is key to helping raise awareness of end user issues to the wider industry. The EUC and the SNIA will continue to collaborate on shared storage industry requirements and goals, and the groups will move forward together to meet the needs of the end user community.

"The EUC annual survey is an important part of our service to the industry," said Norman Owens, EUC Governing Board member. "We appreciate the candor of our end user respondents as we examine the real-world issues."

"At the SNIA, we fully support this open dialogue with the end user community, and we value the direct input it provides," says Vincent Franceschini, chair, SNIA Board of Directors. "As new and improved technologies have emerged, new business and IT challenges have also arisen for end users. IT professionals continue to embrace and leverage new technologies and best practices, but they are looking for improved interoperability, solution integration and ease of use to assist in addressing business priorities. The SNIA recognizes that data storage solutions need to become more integrated into the larger data center framework so that IT can improve its information-based management practices. In 2007, the SNIA is embarking upon new opportunities in technical and educational activities, which will address many of the issues raised in this survey."

The 2007 EUC Survey is now underway, and revisits end user's "top ten pain points" - three years after the initial Survey in 2004. Visit www.snia.org/euc to take this year's survey.