This chapter is from the book
This chapter is from the book
This chapter is from the book
Utilization and Yield
A fundamental piece of the storage TCO equation is utilization and its direct correlation to what can be referred to as the storage yield. If one assumes that the average company used at best 50 percent of their storage assets between 1999 and 2002 (which is itself a conservative number), then, based on the worldwide revenues shown in Table 1-2, we can estimate that over $35 billion dollars in storage assets went unutilized during that time.
NOTE
In this section, I borrow two terms from different fields—the COPQ (from Total Quality Management and Six-Sigma) and yield (from manufacturing and agriculture)—and I apply those terms to the discussion of storage utilization.
Storage utilization is the most important storage management issue today: Poor utilization wastes millions of dollars a year in unused storage assets.
Understanding utilization is crucial for the introduction of ROI, Net Present Value (NPV), and TCO in Chapter 3, "Building a Value Case Using Financial Metrics." This material is required to build the financial models with which the business case for storage networks can be justified.
A close analysis of storage yield and the COPQ demonstrates how increased utilization helps lower the overall storage TCO.
The Cost of Poor Quality and the Storage Problem
The Cost of Poor Quality, in terms of quality and yield management, typically refers to the costs associated with poor or undesirable performance of a product over the course of its economic usefulness.26
A high COPQ implies higher manufacturing, operations, and labor costs, and consequently, lower revenues. Couching the value of an IT solution in terms of quality management, the COPQ can be said to be the dollar value of how a product, service, or solution performs relative to its expectations. In terms of financial analysis, this figure equates to a negative ROI.
Just as the buildup of IT capacity and subsequent downturn was the outcome of macroeconomic events, the move to storage networks is part of many corporations' efforts to raise their storage yield over time and lower the COPQ (and the TCO) for their storage infrastructure.
Storage Yield
In manufacturing operations, the term yield refers to the ratio of good output to gross output.27 In storage operations as in manufacturing, the yield is never be 100 percent as there is always be some waste. The goal of a storage vision is to increase not only storage yields, which can be measured in dollars or percent of labor, but also to increase operational yields (or "good output") as much as possible. Ultimately, a storage vision built on a storage utility model helps increase a company's storage yield, the amount of storage capacity allocated and then used efficiently to create and sustain business value.
A tiered storage infrastructure is required to fully increase storage yield and gain true economies of scale. In Table 1-5, each tier has a different capability model and different direct and indirect costs associated with it. The goal is for the COPQ to be as insignificant as possible (shown here as a percentage of $1,000,000 in revenue), and ideally for the accompanying tiers to be appropriately matched to the level of business impact or business revenue of the associated applications. A typical tiered storage infrastructure might look something like this:
Tier One—Mirrored, redundant storage devices with local and remote replication
Tier Two—RAID-protected, non-redundant storage devices with multiple paths
Tier Three—Non-protected, non-redundant, near-line storage devices (for example, SATA drives used as a tape replacement)
Table 1-6 Cost of Poor Quality as a Percentage of $1,000,000 of Revenue for 1000 GB
|
Storage Type |
Cost per MB |
GB |
Total Cost |
Allocated |
Utilized |
|
Tier 1 |
$0.05 |
1000 |
$51,200 |
80% |
75% |
|
Tier 2 |
$0.03 |
1000 |
$30,720 |
80% |
60% |
|
Tier 3 |
$0.01 |
1000 |
$10,240 |
80% |
90% |
|
Storage Type |
Allocated Yield |
Utilized Yield |
Realized Yield |
Cost of Poor Quality |
COPQ % of Revenue |
|
Tier 1 |
$40,960 |
$30,720 |
60% |
$20,480 |
2.05% |
|
Tier 2 |
$24,576 |
$14,746 |
48% |
$15,974 |
1.60% |
|
Tier 3 |
$8192 |
$7373 |
72% |
$2867 |
0.29% |
As seen, a low storage yield has a corresponding high COPQ and indicates an overall higher total cost of storage ownership. A more complete discussion of tiered storage solutions (and Information Lifecycle Management) is presented in Chapter 5.
NOTE
The difference between allocated and utilized storage is discussed in the section titled "Utilization."
Obstacles Inherent in DAS
As the predominant storage architecture to date in terms of terabytes deployed, DAS has served the storage needs for millions of environments around the globe. Small Computer Systems Interface (SCSI), DAS is a standard, reliable method of presenting disk to hosts. DAS also presents many challenges to the end user including failover and distance limitations, as well as the increased expense associated with poor utilization.
Failover Limitations
Although some DAS environments are Fibre Channel, large storage environments in open systems datacenters have historically been direct-attached SCSI. SCSI is a mainstream technology that has worked well and has been widely available since the early 1980s. SCSI provided the necessary throughput and was robust enough to get the job done. One disadvantage, however, has always been the inability of the UNIX operating system and most databases to tolerate disruptions in SCSI signals, thus limiting the capability to failover from one path to another without impact to the host. In addition, logical unit number (LUN) assignments are typically loaded into the UNIX kernel when the system is booted up, requiring allocation or de-allocation of storage from the host to be planned during an outage window. If the storage unit in question is shared between different clients with mismatched service-level agreements and different maintenance windows, then negotiating an outage window quickly becomes a hopelessly Sisyphean task.
Distance Limitations
Another significant factor hampering the flexibility of SCSI DAS is that SCSI is limited in its capability to transfer data over significant distances. High Voltage Differential (HVD) SCSI can carry data only up to 25 meters without the aid of SCSI extenders. This limitation presents difficulties for applications requiring long-distance transfer, whether for the purposes of disaster recovery planning, application latency, or just for the more physical logistics of datacenter planning.
Expense
Aside from the technical limitations of DAS, the primary drawback of DAS is, without a doubt, its expense. Ultimately, the storage frames themselves constitute a single point of failure, and to build redundancy into direct-attached systems, it is often necessary to mirror the entire frame, thereby doubling the capital costs of implementation and increasing the management overhead (and datacenter space) required to support the environment.
The expense of DAS also stems from poor utilization rates. A closer look at the two primary types of storage utilization further illustrates the nature of the cost savings inherent in networked storage solutions.
Utilization
When considering the impact of managing storage in general and the financial disadvantages of DAS in particular, the primary variable to monitor is storage utilization. Poor utilization leads to a decreased storage yield and a high COPQ, whereby the storage capital asset purchased to provide a service, fails to perform at an optimal level.
Storage utilization has been a marketing hot-button since Fibre Channel SANs began to gain momentum, and as such, utilization is now laden with many different meanings, all of which are often (and unfortunately) used interchangeably. To prevent further confusion, I prefer to use Jon William Toigo's terminology of efficiencies. Toigo clearly delineates between "allocation efficiency" (broadly referred to as "utilization") and "utilization efficiency," which typically reflects the effects of storage usage policies.28
Allocation Efficiency
Due to the physical constraints of the solution, DAS environments are intrinsically susceptible to low "allocation efficiency" rates that cost firms money in terms of unallocated or wasted storage. Let us look at one example of the financial impact of poor allocation efficiency.
Imagine a disk storage system (containing 96 73-GB disk drives) with six four-port SCSI (or Fibre Channel) adapters capable of supporting up to 24 single-path host connections. This system is capable of providing approximately 7008 GB of raw storage, or 3504 GB mirrored. Under most circumstances, hosts have at least two paths to disk, so this particular environment can support a maximum of twelve hosts. In a typical scenario, shown in Figure 1-5, this frame hosts the storage for a small server farm of six clustered hosts (12 nodes).
)
Figure
1-5 Sample DAS Configuration
If each cluster hosts six similar applications using 500 GB each (an allocation efficiency rate of 85 percent), almost 500 GB remains unallocated due to the frame's port limitations. With a purchase price of $0.10 per MB, or $100 per GB, there is a loss of $105,120.00 associated with the unutilized disk on that frame.
NOTE
Keep in mind that as this frame is formatted for mirroring, the value of the unallocated storage is the cost of the total non-mirrored storage. In other words, the 500 GB of unallocated storage is still 1 TB raw, which must be valued at its purchase price. Also note that use the $0.10 per MB for quick math. The average purchase price of the disk might be significantly lower.
This loss can be considered the COPQ and reflects the costs of additional storage required to provide the expected capacity. An allocation efficiency rate of 85 percent for a DAS environment, however, is significantly higher than the normal average. In June, 2001, a joint study published by McKinsey & Company and Merrill Lynch's Technology Group, titled "The Storage Report—Customer Perspectives & Industry Evolution," estimated the average utilization rate for DAS environments to be 50 percent.29
Fred Moore of Horison Information Strategies has an even more dismal view of allocation efficiency. According to Moore, surveys of clients across various industries indicate allocation efficiencies of 30–40 percent for UNIX and Linux environments and even less for Windows environments, which Moore says frequently see allocation efficiency rates as low as 20 percent.30
Using the same environment shown in Figure 1-5 as an example, if the allocation efficiency is only 50 percent, then the loss widens significantly to $350,400, or half the purchase price of the frame. Figure 1-6 shows the costs associated with poor utilization in this environment.
)
Figure
1-6 Utilization Rate and Associated Costs—Cash Basis
Most firms depreciate the cost of storage over the course of its useful life (assuming the storage is purchased and not leased), so the actual COPQ might vary according to depreciation schedules.
Given the rapid progress of technological advancement, in most cases, depreciation is carried out over three years. If the straight-line method of depreciation is used over a period of three years, the asset value or purchase price of the frame is divided by three with the assumption that one-third of its usefulness is consumed each year. The impact of the loss, or the COPQ, is then spread across the span of the economic usefulness of the asset. In other words, one third of the COPQ affects the firm's bottom line each year.
Low utilization does not increase or decrease the estimated life of the hardware, nor does this loss change the asset's value in accounting terms. Low utilization does, however, decrease the storage yield of the asset and increases the COPQ, which, in turn, increases the overall TCO. Regardless of the method of depreciation used, poor utilization detracts from the firm's bottom line.
Whether or not the storage units themselves are depreciated, the net effects of poor allocation efficiency are similar: Low allocation efficiency increases the rate of frequency of additional storage purchases. A real life parallel is buying a full tank of gas and being able to use only half of the purchased fuel. As long as you need to drive the car, you will need to purchase more fuel. If more fuel is not consumed, you will be forced to stop at the gas station more often.
Similarly, as long as the firm operates, it needs to purchase storage. The idea that a firm can delay purchasing storage indefinitely by constantly increasing the utilization rate is, to put it bluntly, misinformed. The long-term key to financial success in terms of storage management is optimizing storage usage to minimize the frequency and magnitude of storage purchases. A high allocation efficiency rate helps decrease the size and number of storage purchases, as does a high utilization efficiency rate.
NOTE
Capacity on-demand programs are alternative procurement strategies aimed at alleviating the frequency and number of storage purchases. Although these "pay-as-you-go" methods are quite successful at easing the purchase and planning process, they do little to address the rate of consumption or poor utilization found in many environments.
Utilization Efficiency
There might be environments in which the allocation efficiency is at a desirable rate, but the allocated storage is misused, unusable, abandoned, or even hoarded. This is what Toigo refers to as poor utilization efficiency, whereby the storage itself might be highly allocated, but poorly utilized. In fact, in many open systems environments in which the storage capacity is efficiently allocated, utilization efficiency might be extremely poor, with many applications needlessly consuming data that is rarely, if ever, used.
To resolve these types of issues, a targeted program or project aimed at reclaiming allocated—but lost or poorly used storage—is needed. A project of this magnitude requires a significant time investment and an energetic executive sponsor who is capable of ensuring the proper alignment of goals and initiatives. A storage reclamation project also requires extensive use of a combination of off-the-shelf storage resource management (SRM) software and home-grown scripts dedicated to tracking storage consumption.
Despite the many obstacles that are known factors in implementing DAS, the majority of disk units sold in the last five years are still connected to hosts in a direct-attached fashion. Most companies—even early adopters of storage networking technologies—are still in the implementation phase of building SANs, and therefore have at least a partial mix of SAN and DAS technologies in the datacenter.
Although it is difficult to determine the exact percentage of DAS and SAN storage currently installed world-wide, estimates based on the sales of disk and Fibre Channel gear indicate that nearly three quarters of all disk storage units installed still utilize the direct-attached architecture. As shown in Table 1-1, DAS storage units made up nearly 70 percent of all storage sales in 2003 (with NAS and SAN storage together comprising approximately 30 percent). As these figures indicate, there is still a long way to go before the majority of storage environments currently deployed are networked storage solutions.
In addition to the recently installed DAS, a mountain of DAS that was purchased during the market upswing and it still carries a sizable net book value. As shown in Table 1-1, nearly one million DAS units were shipped between 2001 and 2003, indicating significant depreciation expense for customers when considering the corresponding low utilization rate (and the high COPQ) for DAS.