Storage Peer Incite: Notes from Wikibon’s May 15, 2007 Research Meeting
This week, Dave Vellante presents The skinny on Hitachi's thin provisioning. Hitachi Data Systems recently announced the Universal Storage Platform V. Lots of performance, but the real story of this announcement was it marks the first leading enterprise storage vendor to integrate delivery-side and services-side virtualization in a single controller architecture. This capability has the potential to reclaim up to 40% of wasted disk space over time; but there are some challenges to adoption that users need to consider.
Over the last number of years, the storage community has waited expectedly for viable enterprise thin provisioning and related virtualization technologies to become integrated and generally available. The forces behind this anticipation are drawn from multiple locations including: (1) end users' growing expectations that the acquisition of storage should be as simple and transparent as using Google Gmail; (2) demands by CFO's to begin gaining greater control over the rate of storage purchases; and (3) storage administrator desires to find better ways to manage multiple different forms of storage and storage resources.
We've noted in the past that virtualization has become an umbrella term for describing the types of technologies that organizations ultimately will flock to over the next 4-5 years as innovations become more available. This past week (5/14/2007) one of the 'big boys,' Hitachi Data Systems, announced the Universal Storage Platform (USP V) which provides one of the first viable examples of how one type of virtualization technology called thin-provisioning can be employed within organizations to change the way storage is administered, purchased and used.
While the USP V has, as any product does initially, certain constraints to deployment --for example only resources behind the USP V are available to be controlled by the USP V and the new function is not supported on older USP's as a software-only upgrade-- we note that this is an example of a product that integrates both backend data and storage virtualization with front end storage administration and application services virtualization. Over the next few months, we expect users will be very focused on exploring how the USP V platform can start to deliver benefit in their organizations.
We think they'll find several obvious answers. First, USP V and technologies like USP V may re-claim as much as 20 - 40 percent of dead storage in many application domains. Secondly, that these technologies can dramatically improve the mechanisms that IT and storage administrators utilize to communicate needs, delivery and value to their businesses. And third, that these technologies will more fully integrate the way the marketplace is beginning to look at storage as driven by more consumer-oriented organizations: namely, that storage is best perceived as a service tied to application requirements and not as hardware to be purchased, provisioned and allocated years in advance of its actual use.
Today, $1 or so per GB is going to added to the USP V overall cost of storage (Thin provisioning added costs), but as competition from EMC, IBM and other new startups begins to heat up around the thin provisioning market, we expect that price point as well as existing constraints on accessing all forms of storage resources from this very powerful type of capability will start to pick up dramatically.
Action Item: User organizations should view the USP V as an excellent example of how thin provisioning technologies will be packaged and put forward into the marketplace over the next few years. Near-term, new and Tier 2/Tier 3 applications should gain some significant economic benefits from improved capacity utilization. We expect this technology will become the centerpiece of discussion for all users and suppliers of storage over the course of the next few years.
Thin provisioning promises to offer IT both substantial benefits in terms of reclaiming wasted space, and critical support for the development of new business models that are more in line with those of consumer storage service suppliers. The Hitachi USPV represents the first of many enterprise scale innovations to come in this space.
Questions remain regarding how to get started with thin provisioning. The following can serve as userul guidelines:
- Start by understanding the benefits of thin provisioning. Specifically, what are current storage utilization rates, how much capacity is being allocated (wasted) to accommodate future growth and how much space can be reclaimed? Ranges of 20 - 40% reclamation (in utilization percentage points) are by no means out of the question but organizations need to 'personalize' this analysis.
- Fully understand the constraints to adoption. Specifically, what skills, application knowledge, process changes and business terms need to be developed to exploit thin provisioning.
- Prioritize the applications that are the best candidates for thin provisioning by using a combination of economic justification and degree of difficulty.
Ironically, while the best targets to start thin provisioning are likely to be 'safer' Tier 2 and Tier 3 applications, the USPV doesn't currently support thin provisioning on devices external to the controller-- the best candidates for T2 and T3 data sets. Does this mean users should wait? Perhaps, but more often than not, Hitachi's customers have T2 and T3 storage that resides internal to the USP and they can use Hitachi's virtual partition manager to support that tiering. Users should note that thin provisioning is not supported on existing USP's and they are advised to place T2/T3 data on USPV's in order to get up the thin provisioning learning curve with less risk.
Action Item: Users should aggressively assess thin provisioning for enteprise storage systems. Start with T2 and T3 applications and understand the costs, application nuances, process changes and ISV support offered. Use the next six to nine months as a learning runway which will speed adoption and time-to-benefit when the next round of announcements hits the market.
The best technology buys are ones that pay for themselves – fast – and promise an unambiguous stream of very good future returns for both the business and the IT organization. Virtualization using thin provisioning looks like one of the best technology buys in a long time. First, simply by reclaiming dead storage, the technology should be able to rapidly improve the space productivity of disk by up to 40% without any appreciable impact on performance. Second, by virtualizing both the delivery (disk-to-controller) and service (controller-to-application) sides of storage, the technology can catalyze much more effective allocation procedures and overall working relationships within IT (infrastructure to application development) and between IT and business (applications operations to business operations). A thorough assessment of the HDS UPS V – the first comprehensive implementation of thin provisioning technology from one of storage’s "big boys" – should be on the "to-do" list of every enterprise storage group.
Action Item: Storage administrators should carefully study the applicability of the HDS USP V to their business’s storage needs and get the results in front of their CIO, who will appreciate finally having a great storage financials story to tell to CFOs while setting up overall service improvements to application owners.
Thin provisioning is one of the few storage advances where the concept and benefits can be understood by a layperson in less than a minute. However, if thin provisioning is implemented just to solve over allocation of storage, the main potential value will not be realized. Oracle has shown how the integration of its 10g database software with virtual volumes, automated placement and thin provisioning can make systems simpler to setup, grow, and monitor. Oracle uses Auto Extend and its Automatic Storage Manager to utilize the advanced functionality of a storage controller to provide storage services. This type of approach should be a model for software and internal development.
In addition, users should note that clearly some software is not well-suited to thin provisioning. Microsoft's NTFS for example prefers new, unused blocks rather than reusing released blocks unless it is formatted the appropriate way. Moreover, certain other applications will grab all storage allocated and write metadata throughout the space. If that space is not physically available, applications will crash.
There are workarounds for most of these problems in the short term that typically involve configuring an application in a way that disables certain functionality so that it won't conflict with the way thin provisioning operates. Users should obviously assess the potential tradeoffs of configuring applications in this manner against the economic benefits of thin provisioning.
Mid-to-longer-term, as thin provisioning is aggressively adopted, software vendors will embrace thin provisioning because it makes their software easier to manage and cheaper to provision. Larger customers, storage vendors in general and Hitachi in particular should work aggressively with independent software vendors (ISV's) to agree best practice communication between applications and storage in order to accelerate the adoption of thin provisioning. ISV's, storage vendors and customers win with this type of collaboration.
Action item: IT should work aggressively to implement a new way of connecting applications, storage and budget owners. Early adoption of thin provisioning should focus on tier two and tier three applications and on ensuring that the right reporting and business processes are in place to communicate allocated, actual and projected use of storage resources.
When 3PAR did a major raise of capital, just before 9/11/2001, it wasn't at all clear to most observers that thin provisioning had game-changing potential. Indeed, many thought prospects would dwindle along with CIO buying patterns of late 2001 through 2003. With the announcement of the Hitachi USPV, however, it is becoming more obvious that thin provisioning is here to stay.
While smaller vendors and specialized products have pioneered thin provisioning for several years now, often in file-based environments, it is clear that thin provisioning as applied to block-based storage holds great promise for users. In this regard, Hitachi Data Systems has a lead over other large enterprise storage companies that most likely ranges from 9-12 months.
Simply put, vendors must respond to this announcement or they will be left out of a major storage wave over the next 3-5 years. HDS is not sitting still as evidenced on a Wikibon.org research call where the company indicated that external thin provisioning can technically be done today on the USPV (with Universal Volume Manager) and will be generally available later in 2007. Both EMC and IBM will have to offer competitive responses in the next twelve months; other smaller less established players must be more aggressive or they'll be left behind.
Action Item: Vendors of high end storage subsystems should agressively introduce thin provisioning. HDS may in fact enhance its thin provisioning functionality (providing support for external devices) prior to EMC and IBM announcing the capabiity for block-oriented storage. Key to adoption will be ISV support of thin provisioning, an area that needs more attention by established players.
The availability of thin provisioning on just the new Hitachi USP models makes it harder to plan for thin provisioning except for larger or high growth installations. For these data centers, a new USP V controller can be introduced as part of natural growth, and thin provisioning introduced to help accelerate its adoption.
Assuming that Hitachi enables thin provisioning on external storage later in the year, new and large customers will have some exciting opportunities to reuse older arrays and/or use low cost arrays and have the new USP V provide the thin provisioning for them. Storage administrators should remember that drives follow a “U” shaped curve in terms of mean time between failure (MTBF), meaning drive problems often occur early in a lifecycle, lessen in middle year and increase again as drives age. As they age like baby boomers, administrators should configure and allocate these devices to minimize unavailability.
Smaller and low growth USP customers should pressure Hitachi to think like a software company and announce thin provisioning on current versions of the USP. Wikibon encourages the community to contribute possible migration strategies from one generation of USP to the next.
Action item: Where and if possible, plan for the aggressive use of the USP to apply thin provisioning to existing arrays as externally attached storage. The business case in most instances will be compelling.