Although the cost of disk storage has been declining continually, the demand for storage is ever increasing. Therefore, good storage management practices are still an important discipline for data center managers. However, data center benchmarks collected by Metrics Based Assessments, LLC show that disk storage management practices have declined in the past decade.
The current benchmarks shown in this article are taken from the book, Best Practices and Benchmarks in the Data Center, which is available for purchase on our website.
Measuring Storage Management Practices
One measure of effective disk storage management is the percentage of storage that is unused. Although data centers need some amount of unused disk space to accommodate growth in data volume and spikes in transactional business, having too much unused disk space is simply a waste of money. With the cost of disk storage declining every year, it does not make sense to have excessive storage capacity far in advance of need.
Another measure of good storage management is to look at the percentage of disk space that has been accessed within some period of time (e.g. 15 days). A disciplined approach to disk management considers how often files are accessed and what kind of response time is required for retrieval. Those files that are infrequently accessed and do not require immediate retrieval are best archived to offline storage, reducing online storage requirements.
Unfortunately, by both of these measures, disk storage management practices have declined over the past decade, in mainframe, UNIX, and Windows server environments.
Mainframe Storage Management Performance
As shown in Figure 1, management of disk storage in mainframe environments has declined since 1997, by both measures of storage management discipline. In 1997, the average benchmark participant in the mainframe environment had 68.5% of disk space allocated with 575 installed MIPS and 2.2 TB of installed disk storage, leaving less than a terabyte unused. Today, nine years later, the average mainframe benchmark participant has grown to 2,208 installed MIPS and 11.6 TB of disk installed, but only 61.3% of it allocated. Therefore, 4.5 TB of installed disk is empty.
In other words, in the average mainframe environment today, the amount of disk space unused is twice the amount of disk space installed in 1997.
By the measure of data access, management of disk storage is also declining. In 1997, 78.8% of allocated space was accessed in the previous 15 days. Today only 72.6% of allocated space is accessed 15 days, as shown in Figure 1.
Why have mainframe storage management practices been declining? There are several reasons:
- The disk management function in many data centers is understaffed. Fewer storage managers mean less storage management. In 1997, there were approximately two storage managers per terabyte managed. Today, there are only approximately 0.5 storage managers per terabyte.
- Even though mainframe systems have excellent tools for storage management, in many shops these tools are not fully implemented. The average percent of installed disk space controlled by System Managed Storage (SMS) is only 61.7%.
- Capacity planning is understaffed at many mainframe shops. In 1997, there were 0.0083 capacity planners per used MIPS. Today, there are 0.0013 capacity planners per used MIPS.
The primary cause for declining disciplines in storage management, however, is the falling cost of mainframe disk storage. As disk space becomes cheaper, data center managers have less incentive to pay attention to efficient storage management. In 1997, the average annual cost per installed gigabyte was $1,136. Today this metric has declined to $90, with some organizations paying as little as $10 per installed gigabyte. The falling cost of storage easily hides the inefficiencies in storage management that are creeping in to many data centers.
UNIX and Windows Storage Management Performance
When we examine UNIX and Windows environments, we find that disk storage management practices are even worse than with mainframe environments, with one exception. When storage area networks (SANs) are used to manage storage, the management practices are better, as shown in Figure 2.
UNIX and Windows environments manage storage as a combination of locally attached storage and SANs. The percentage of storage managed by SANs is nearly the same for UNIX (65.1%) and Windows environments (63.6%). As shown in Figure 2, UNIX environments average only 56.6% utilization for locally attached storage, but this metric rises to 75.5% for storage managed by SANs. In Windows environments, however, storage management performance is poorer, with only 46.6% of locally attached storage allocated and only 55.8% allocated when SANs are deployed.
Storage Management Staffing Ratios
One final measure of the attention given to storage management is to look at the ratio of disk storage to storage management staff. In other words, what is the average amount of disk space managed by each storage manager? As shown in Figure 3, storage managers in mainframe environments are responsible for 2.2 TB of disk space, on average, while in UNIX environments they must manage 9.5 TB and in Windows environments they are responsible for over 144.9 TB of disk storage.
The figure for Windows environments is especially striking. Since most Windows environments have significantly less than 144 TB of storage, this means that most Windows shops have no one whose primary job is to manage storage.
As we have seen, disciplines in managing disk storage are not improving. In mainframe environments, where storage management has historically been the strongest, storage performance has been falling by every measure. Disciplines in UNIX environments, and especially in Windows server farms, are even worse. The falling cost of storage may seem to make storage performance and utilization a lower priority. However, the amount of data being managed in many organizations is exploding. Therefore, effective storage management is still needed to control the costs and optimize the performance of storage. Although SANs and storage management tools can help in the effort, ultimately there must be skilled staff members whose job it is to manage this critical data center resource.
This article is based on metrics provided by Mark Levin, a Partner at Metrics Based Assessments, LLC, from data collected from thousands of data center benchmarking studies over the past 16 years.
Current benchmark data is available in Levin’s book, Best Practices and Benchmarks in the Data Center.