The Business Case for Storage Networks [Electronic resources] نسخه متنی

Cost-Saving Strategies for Storage-Centric Firms

Becoming a storage-centric entity requires executive-level sponsorship and active participation of individual contributors who foster awareness at every level of the organization of the value of storage as a depreciating capital asset with a corresponding yield and impact to the company's bottom line.

Reiterating the premise from earlier chapters, it is possible to drive significant change in an organization, from the bottom up as it were, by focusing on storage as an asset. Much like an EVA organization, a storage-centric organization views IT decisions with a focus on storage management strategies to ensure that value is created or at minimum not destroyed. Storage consolidation, server consolidation, virtualization, and Information Lifecycle Managementkey initiatives in every storage-centric organizationcan all be driven from the bottom up, given a viable, cohesive storage vision and active, executive-level support.

Storage Consolidation

Chapter 3, "Building a Value Case Using Financial Metrics," analyzed the financial benefits of a consolidation project, whereby Goodrich was able to eliminate $5,760,000 in maintenance fees by consolidating 80 external storage frames. The savings from maintenance fees were augmented by space savings in the datacenter, which deferred a $4,000,000 datacenter expansion project. In addition to resulting in fewer points of management, the consolidation effort provided a centralized agenda for lowering costs.

Consolidation is, of course, not without its caveats. The process of moving terabytes of critical data storage from multiple external arrays to just a handful does require significant foresight and planning to ensure that all high-availability requirements are met. A single, extended outage for several hosts attached to a 10- or 20-terabyte frame can affect large portions of a Fortune 500 company and virtually wipe out a small-to-medium sized business. Therefore, when planning a large-scale consolidation, it is critical that considerations be made for high availability and recoverability, and that the affected clients are aware of (and agree to accept) the possible risks.

Note

As consolidation begins to gain traction, environments are disturbed and many applications that had been forgotten are discovered. These applications might have less stringent requirements since their original deployment or they might not even be in use, at which point, the server and application can be fully decommissioned.

The consolidation bandwagon will likely become something that disparate business functions want to get behind. Other teams will want to take part in the success of storage consolidation and will either support the initiative or start separate consolidation initiatives. Application teams might respond positively to a bounty system for recovery, whereby rewards are given for the most terabytes recovered or decommissioned. Business units might offer similar rewards for applications decommissioned.

A storage-centric business strategy built on consolidation is capable of driving significant change in an organization.

Consolidation, a process marked by repeated planned outages to install HBAs and to copy databases, also requires considerable effort on the part of storage managers to ensure that disks are properly allocated and utilized. The process of dynamically migrating data to increase utilization and bolster operational efficiency is one that storage software manufacturers seek to automate.

Server Consolidation

Chapter 1, "Industry Landscape: Storage Costs and Consumption").

In either case, the potential to provide increased availability to more applications at a lower price point is there. Regardless of the decision to implement fewer high-end server platforms or more low-cost servers, the decision to increase the application-server ratio is one that must be seriously considered to lower TCO over the long term.

The issue of risk versus reward raises its head in any discussion of consolidation, and server or application consolidation is no exception. The basic argument for server consolidation is easy to understand: Due to increased processor power, better fault management and load-sharing software, and lower costs, it is now possible to relieve the burden of expensive hardware maintenance costs by collocating applications on the same hardware. As long as the performance, availability, and uptime requirements can be met, then server consolidation, either to Linux farms or high-end servers, should be transparent to the application owners, with the exception of the downtime required to move the application.

Application consolidation is a riskier proposition and one that is harder to sell based on perceived diminished rewards. There are tangible benefits from using fewer software licenses required to perform the same tasks; however, despite the benefit of decreased points of management, the process of migrating and collapsing applications is much harder for businesses and IT departments to agree to than storage or server consolidation. Consolidation at the software level addresses the intricate methods that business functions use to interact with IT and with each other. Consolidation at the hardware level makes sense to many individuals because of the nature of computing advancement (Moore's Law) and the concept of an asset's useful life. Application infrastructure, however, reflects a company's proprietary knowledge whose wealth and value go far beyond that shown on the balance sheet. The primary issue of application consolidation is a question of business process engineering and requires a much broader scope of involvement across the enterprise.

This is not to say that application consolidation should be dismissed as a potential opportunity for lowering costs, but only that it should not be entered into lightly. On the one hand, the gains from a large-scale application consolidation effort can be significant, but it takes time, energy, and focus to turn them into a reality. Server and storage consolidation, on the other hand, can be sold as quick wins, which might spark some interest and initiative in consolidating applications. Consolidation at the disk, server, and application level ensures that the firm is capable of increasing the utilization of its assets.

Virtualization extends the concept of increasing utilization to the firm's entire set of computing assets. The ultimate goal of managing any resource is to achieve its maximum utilization rate. This applies not only to enterprise disk assets, but also to server and application entities. Unless there is a compelling reason to maintain a buffer of unutilized disks (or CPUs or switch ports), underutilization signifies waste. Tools designed to virtualize resources greatly simplify the processes behind management and consolidation, and therefore increase the utilization of those assets.

Virtualization

Although virtualization software packages are in the early adopter phase, the promise of virtualization of the CPU and the disk coupled with the capability to eventually shield system owners from storage and system administration pain (while increasing utilization) appeals to decision makers.

Host-based virtualization products have been in use in production datacenters for over a decade. Applications, such as VERITAS Volume Manager and Hewlett-Packard's Logical Volume Manager, provide transparency between the host and the storage unit to simplify the management of thousands of logical devices. These types of disk virtualization products are widely accepted solutions.

Network-based virtualization provides an additional layer of abstraction between heterogeneous storage and the hosts on a storage network, which eases the management of different storage platforms across the network. Virtualization at the network level increases application uptime by allowing resources to be dynamically allocated in the event of a planned or unplanned outage. Network-based virtualization also increases allocation efficiency rates by allowing devices anywhere on the storage network to be reassigned without impact to the application or the end user.

Companies, such as IBM and VERITAS, the first to bring to market intelligent virtualization solutions embedded on a switch, have set the pace of development with their releases of SAN Volume Controller and Storage Foundation for Networks respectively, and they will quickly erect barriers to entry to prevent further competition. Other types of virtualization products will quickly come to market to meet pent-up demand for virtualization functionality.

Early adopters of virtualization products find performance, reliability, and interoperability issues to be a factor, but for large environments in which the potential benefits of virtualization far outweigh the risks and costs associated with implementing immature products, virtualization is already making inroads.

As networked-based disk virtualization products mature, labor costs for managing storage across the enterprise decrease dramatically. Likewise, the virtualization of the CPU decreases the TCO for servers and applications. In addition to IBM and VERITAS, whose virtualization solutions are both available as separate service modules on the Cisco MDS platform switches, MonoSphere and Egenera are two more companies whose products are designed to virtualize and optimize computing assets.

MonoSphere

MonoSphere was founded two and a half years ago on the premise that corporate leaders facing dramatic growth in data storage would soon recognize the strategic importance of managing storage at the enterprise level and the need to address the rising TCO associated with managing heterogeneous storage with fewer staff.

Note

In February, 2004, I visited the corporate headquarters of MonoSphere, makers of cross-platform automated storage management (ASM) software solutions, to gain some insight on the storage software market, and to understand the effect MonoSphere's leadership believes virtualization and consolidation will have on the overall market for storage hardware. When I was at MonoSphere, I spoke with Ray Villeneuve, President and CEO, and Shridar Subramanian, Director of Business Strategy and Alliances.

Similar to CreekPath, Onaro, and other storage-related independent software vendors (ISVs), MonoSphere believes that future growth in the enterprise storage market is heavily tied to advances in software designed to manage and leverage networked storage. MonoSphere differentiates itself, however, from other storage software providers with the scale and scope of its flagship product, MonoSphere Storage Manager™, which is designed specifically to automate tasks and policies that increase utilization of and lower the TCO for enterprise storage assets.

In April, 2003, the company began shipping MonoSphere Storage Manager™ for the Windows platform and, later in 2003, they shipped versions of the product for Solaris and Linux hosts.

MonoSphere is headquartered in Silicon Valley and has a research and development center in Tel Aviv, Israel.

The MonoSphere Storage Manager™ product is designed to provide a highly detailed view of the utilization of an enterprise's storage assets and to provide a methodology and toolset for migrating data to unused or unallocated storage, thereby increasing allocation efficiency and protecting the value of the storage asset.

MonoSphere Storage Manager™ virtualizes storage devices and allows the user to create virtual pools that can be allocated and de-allocated with little or no impact to the host environment. This abstraction of the storage device, as seen by the host, facilitates the use of a tiered storage strategy to lower the overall total cost of storage ownership. The simplification of implementing a tiered approach to storage helps customers avoid some of the interoperability issues associated with heterogeneous storage, thereby increasing purchasing power for storage decision makers.

From a single, out-of-band console on the storage network, storage managers can view reports showing true capacity utilization based on a fine-grained view of which storage blocks actually contain user data. This is in contrast with SRM products that cannot distinguish between storage that is allocated and that which is used. To act upon insights gleaned from these reports, storage managers either use prepackaged (canned) policies or they create their own policy-based rule sets to manage data layout across pools of free storage.

At the application server level, a software driver continuously monitors usage patterns, latency, and throughput, providing historical trending and assisting the environment's owner with ongoing policy refinement.

Intermediate volumes, known as MonoSphere Adaptive Volumes, are used to stage and destage data as it is moved between pools of storage by a separate, in-band server dedicated to the MonoSphere Storage Manager™ application. This server handles the actual data migrations between the unused devices, pools, and tiers as dictated by policies specific to the environment, and it minimizes the impact of the migrations on the hosts on the storage network. A common application for this unique capability is to create "spillover" storage that is used as local storage capacity becomes filled. In this way, local storage can be made to behave as if it were infinitely large, without impacting applications, so applications never outgrow their storage.

By simplifying the process of data migrations between tiers and by providing an accurate and up-to-date view of an environment's allocation efficiency, MonoSphere Storage Manager™ is positioned to radically change the way storage is managed in today's enterprise.

Egenera

Egenera, based in Marlboro, Massachusetts, intends to capitalize on the confluence of events and market drivers that it believes led to the next inflection point in the server market. Egenera builds the Egenera BladeFrame system, pools of massively scalable processing resources designed to meet the market demand for highly redundant, highly available environments. The nature of the blade computer, coupled with the virtualization of the IO components in the subsystem, means that the BladeFrame can lower the TCO by increasing the utilization of the processor much in the same manner as SANs increase the utilization of disk storage.

Note

In February, 2004, I spoke at length with Susan Davis, Egenera's Vice President of Product Marketing and Management, about server consolidation, grid computing, and the latest inflection point in the market for high-end servers.

Egenera was founded in 2000 by Vern Brownell, former CTO of Goldman Sachs, as an answer to the primarily physical problems he saw facing large enterprise datacenters. The traditional view of the server as an isolated resource, similar in concept to direct-attached storage (DAS), along with the need to provide highly-available computing power, has led to rampant growth of server islands that have become a cost and management nightmare. The proliferation of servers and storage in the datacenter, the increased points of management, and the related costs inspired Brownell to design a system capable of meeting performance and reliability constraints while reducing the number of managed resources in the enterprise. The end result has been a fundamental change in the concept of the server itself.

Server consolidation (or grid computing in a larger sense) is at a basic level about raising the utilization of the processor by providing sharable computing resources as a utility. Just as storage networking and storage consolidation is about increasing the utilization of the storage resources, processing area networks, such as those built by Egenera, increase the utilization of server resources by allowing groups of CPUs to be allocated to and de-allocated from an application's resource pool. The unutilized processors can therefore be shared between environments lacking resources, thereby offering investment protection and increased return on investment (ROI).

The virtualization of IO components in the BladeFrame creates a holistic view of the resources managed with intelligent software to make allocation and assignment of resources a painless process. Whereas in the past, the purchase, installation, and assignment of server resources in the datacenter involved the management of a number of different processes and physical elements, the use of blade farms to virtualize all IO components at the server level adds flexibility while reducing operational complexity.

This level of virtualization goes beyond the common concept of layers of abstraction provided by complex software stacked on top of numerous hardware entities. Egenera seeks to limit the number of IO entities underneath the virtual view provided by its management software.

Susan Davis believes that the inflection point in the server market is the result of the recent economic downturn, increased processor power, the growing acceptance of Linux in the datacenter, and, of course, customers' requirements to reduce acquisition and management costs while providing world-class technology solutions for their clients.

Egenera's successive years of triple-digit growth indicate that even though the grid computing concept is still in the early adopter phase, the market for solutions such as those provided by Egenera is rapidly growing.

Information Lifecycle Management

Instituting an information lifecycle management (ILM) framework, a process designed to guide the migration of data to different tiers or classes of storage as its usefulness declines over time, is a fundamental strategy designed to lower costs. As part of an ILM infrastructure, different tiers of storage with varying levels of functionality and availability are provided at different cost structures. As the value of the data declines over time, it is migrated to cheaper and cheaper disks until it is archived or deleted.

Note

Migration to storage networks and consolidation should be considered a fundamental step before initiating plans for instituting an ILM infrastructure.

As a holistic approach to IT infrastructure, ILM addresses the data itself, the kernel of intellectual property at the firm's core, and it weighs the value of the data against the costs of the infrastructure required to support it. In many ways, ILM extends the core-context debate into the tangible realm of managed terabytes: Core data is that which should reside at the uppermost tier, whereas contextual data can be transformed into a more manageable format or deleted entirely. In either case, it is necessary to determine the break-even point at which it becomes too costly and unwieldy to provide tier-one support for tier-three data. An example of this analysis is provided in the next section.

Note

Also keep in mind that the labor involved in implementing both SRM and ILM in your framework increases your TCO for the solution. Including the costs of the SRM and ILM management solution in the TCO analysis for your environment is required for an accurate portrait of cost trends.

Managing Costs in an ILM Environment

Managing storage costs for the life of the storage asset requires up-to-date and accurate cost data at each tier and across the enterprise. Few organizations have sufficient time and energy to devote to tracking and fine-tuning TCO data. Where possible, TCO initiatives can be driven by the storage team, but they are always reliantat least to some degreeon asset management and purchasing data to provide granularity and accuracy.

To correctly position the data in the requisite tier, it is necessary to analyze the break-even point for managing that data according to its priority, business relevance, and revenue impact. For the purposes of this discussion, we cover only three tiers: a high-end tier (Gold), a mid-range tier (Silver), and a low-end tier (Bronze). Additional tiers can be created, although management costs increase as the number of tiers increases. In the following examples, the per megabyte purchase price is (inclusive of maintenance) $0.10 for Tier 1, $0.05 for Tier 2, and $0.03 for Tier 3. In addition, the following assumptions are made for each tier:

The discount rate is ten percent.

Future labor costs are discounted using net present value (NPV).

One full-time equivalent (FTE) can effectively manage 1.5 TB.

Utilization is ignored.

Growth is stagnant.

The depreciation schedule is three years.

Cash basis is for all three years.

This section shows an example at each tier.

Note

It is critical to note that at this point in time, heterogeneity is key to gaining economies of scale, and management costs through a tiered infrastructure without the benefits of interoperability initiatives such as Bluefin, actually increase with the addition of different platforms into the support matrix. As hardware and software products mature and long-awaited interoperability subsequently materializes, then and only then will a tiered infrastructure be able to aggressively lower management costs.

Currently, hardware solutions for each vendor at each tier (low-end, mid-range, high-end) require different management interfaces and different support processes, hampering the ability to scale support for different platforms across the enterprise. Although products from the major disk vendors have Java-based or web-based GUI interfaces, there is no way at this time to manage, provision, allocate or de-allocate storage across a vast array of different storage solutions from a single console without a third-party product.

The capability for a tiered storage infrastructure to increase supportability for hundreds of terabytes is wholly dependant on hardware interoperability and virtualization software. For the purposes of the tiered TCO discussion, the assumption should be that virtualization software as a mature product has been implemented and that the tiered solutions are transparent to the storage managers and the end usersa scenario that should materialize in production datacenters by 20052006.

Classifying Tiers Based on TCO

It is important to understand that migration from DAS to SAN or network-attached storage solutions (NAS) (and from a non-tiered to a tiered storage environment) invariably requires significant time and resource commitments. These migrations do not happen overnight. During considerable lengths of time, there might be multiple environments in place for the same application at the switch, disk, and host level. These duplicate environments increase the management, and hardware costs contribute to an initially higher TCO. Consolidation efforts also involve the creation of duplicate environments, which raise the hardware cost components associated with the consolidated environments. Over time, the costs decrease; in the short run, costs spike. The following examples should help identify the break-even point of TCO and tier application.

Tier 1 (Gold)

This example assumes the following:

The environment uses a total of 1500 GB (including local and remotely replicated copies).

The $0.10 purchase price includes all licenses and maintenance.

The value of one FTE is $100,000.00.

Installation costs are $0.01 per megabyte.

Backups are taken weekly and twice daily.

Note

Note that for the cash basis analysis, the FTE costs are the NPV of one FTE over three years using a ten percent discount rate.

Table 5-2 shows the TCO for a typical Tier 1 storage environment. Note that this number is the TCO for the storage only and does not include the costs for server hardware or application licenses. Cost components are shown as both cash basis and depreciation basis.