Virtual tape: a solid citizen in an ILM world.

Enterprise storage has rapidly evolved in recent years from a "two-size-fits-all" (high- performance disk and low-cost tape) world into a complex mix of discrete storage elements with a wide variety of performance, cost and capacity attributes. This alphabet soup of technologies (SATA, iSCSI, etc.) presents an unprecedented opportunity for storage managers to locate the right data on the right device at the right time, while satisfying another alphabet soup of business and regulatory requirements (HIPAA, Sarbanes-Oxley, etc.). Moreover, software tools have emerged under the SRM banner to more efficiently manage allocation of storage.

However, these technologies also present a mind-boggling, complex set of management challenges which, if not addressed properly, could negate the potential business benefits associated with what has come to be known as Information Lifecycle Management.

One common assumption underlying the implementation of ILM is that organizations can dramatically reduce cost while improving access to critical data and compliance with stringent regulatory requirements. At the hardware level, this essentially comes down to introducing new layers of disk technology into the storage hierarchy. ILM is, after all, at least partly a variation on the old theme of hierarchical storage management (HSM).

Frequently discussed in this context is some form of ultra low-cost disk, such as Serial ATA. This disk is commonly associated with backup data that is staged for fast recovery or reference data that is required on a "reasonably" timely basis. Each of these applications (backup and reference data) of SATA disk require an application to access data written to this class of storage devices adding some degree of complexity associated with creating and managing yet another layer in a storage hierarchy while also managing the data itself.

In no context has the notion of backup and archive been eliminated from the ILM discussion. In fact, they remain ever more important in the hierarchy, as organizations strive to better manage their growing stores of data and preserve that data into perpetuity.

What if you were able to achieve the desired cost benefits of relatively fast, low-cost intermediate storage without adding the complexity associated with managing it? That is where virtual tape comes in. CentricStor is a heterogeneous virtual tape system designed and manufactured by Fujitsu Siemens Computers and now available in the U.S. through PeakData, Inc. and its reseller partners.

CentricStor maintains the traditional two levels of the hierarchy (primary storage and archive) from a management perspective, but delivers "virtually" two to four performance levels all controlled by management policy within the archive. In addition, it supports duplexing or Export/Import for business continuation. The levels within the archive could be fast cache, slow cache (with binding capabilities), fast tape, capacity tape, WORM tape separate from re-usable tape, and Iron Mountain or the equivalent.

Our analysis of CentricStor within an ILM strategy reveals the potential for significant cost savings when compared with using a three-tiered storage approach. ILM represents the intelligent, automated, migration of data from higher performing storage subsystems (expensive) to lower performing storage subsystems (lower cost) during the useful lifecycle of data with the explicit goal of reducing the cost of storage for all of the data within an enterprise.

A goal of a cohesive ILM strategy is to eliminate server workloads by migrating data through each level of storage (e.g., from primary to secondary storage and from secondary storage to long-term retention). A design requirement for any long-term storage decision (and ILM solution) includes non-disruptive introduction of newer data storage technologies into the migration levels.

Another important benefit of an effective ILM strategy is to reduce storage complexity, thus reducing storage administration costs. ILM solutions will require separately defined migration and recall requirements for each of the various user-defined classes of data (e.g., mission-critical, vital, sensitive, non-critical).

A complete virtual tape system such as CentricStor within ILM can provide the following benefits:

* Simplifying the administration of data once written into archive with assurances that data is always recoverable

* Satisfying various levels of redundancy to assure service-level commitments to Lines-of-Business for data availability

* Leveraging magnetic tape-based data storage solutions to provide the lowest average costs per MB possible

With the addition of an integrated virtual tape 'front end,' a comprehensive tape automation subsystem becomes a viable, compelling participant in the ILM architecture. A comprehensive virtual tape system can bring the acquisition cost of storage below $0.001/MB while also delivering policy-based management and media assurance for long-term data availability.

Integrating a complete virtual tape solution into an ILM-based operation enables backups and ILM-migrations (archiving) to use common resources providing further operational savings in administration. Retention periods are governed by several factors such as regulatory compliance, operational efficiencies, corporate policies, and user requests. Each of these retention periods is controlled by the archiving application that enables access to migrated data (see Figure 1).

If we assume that: (a) data growth is linear, that (b) data actively accessed within 30 days of creation is one half of the data accessed within 30 to 90 days, and that (c) there is a 5-year data retention policy in force--we would calculate that the long-term storage would represent 57 of 60 total months of data created, or 95% of the total storage.

For the sake of simplicity, let's assume that there is one TB of mission critical data in primary storage. Mission-critical data is assumed to be 15% of the online data in primary storage; so, by calculation, the total data within primary storage is 6.7TB. Forty percent of the data is non-critical, leaving 60% to be placed on replicated disk systems, mirrored or perhaps hot failover capable.

In a mirroring disk example there may be two instances of data on each of the failover systems with each instance having additional capacity for growth and management--in effect doubling the available data storage space. One TB of mission-critical storage could require 32TB of primary storage (60% X 6.7TB = 4TB; mirrored X 2 subsystems = 16TB plus additional free space for growth and management = 32TB). If the same disk subsystems were used for secondary storage, an additional 64TB of on-line disk capacity would be required.

The one TB of mission-critical data residing on primary storage (assuming linear growth and mission critical is 15% of the total online) is a small of portion of the total enterprise requirement of 400TB. This example requires a long-term storage capacity equal to 95% of the total 60-month data capacity equating to 380TB. As an archive built upon LTO-2, it would require 950 cartridges. In the archive, ensuring long-term availability, there would be at least four copies of the media (two local and two remote) requiring at least 3800 tapes.

Total storage hardware costs for this example would amount to $6.7M. Distribution of costs across the three tiers of the ILM model are as follows:

* Primary storage consumes 70%

* Secondary storage: 15%

* Long-term archive: 15%

If all primary and secondary storage were deployed on high-performance, redundant, mirrored disk systems, total systems costs would increase significantly.

Secondary storage is the issue--and where the complexity comes in. No one disagrees that mission-critical data needs to be on the very best disk systems available. The question is about the "aging of data" and getting old data out of the way so that expensive resources are available for priority computing tasks.

In our example, we see 13.4TB of secondary data. The likelihood of reference to that data is less than 5% at 30 days. After 90 days, it is perhaps less than 1%. Meanwhile, customers are keeping this data on their most expensive, highest availability disk systems with 2 to 5 mirrored copies. I say: archive aggressively, use CentricStor cache to replace secondary storage and leverage the fact that all data will reside on tape. Using CentricStor in your ILM process means that a single copy of the data in CentricStor cache has the redundancy of multiple copies of data on tape to begin with, thus mirrored vertically, rather than within the same technology.

Savings generated by deploying CentricStor into the ILM architecture are several. Specific hardware savings amount to nearly $1M in the example described above. In addition, there is the elimination of an entire level of data migration resources from server workloads and elimination of storage administration for the 13.4TB of secondary storage equipment.

To provide data backup for primary storage requires resources to move 6.7TB of primary data and, by simple calculation, 13.4TB of secondary storage or 20TB in total. Fully two-thirds of the on-line storage being backed up does not change, but still consumes expensive backup and administrative management. A backup solution for this amount of data would require an investment of $700K for drives, media and libraries.

Aggressive use of CentricStor as the secondary storage resource will save substantially on the backup window. The secondary storage amounts to 13.4TB which would be included in the migration into archive at the end of 30 days, eliminating the backup requirement. The additional cost for backing up primary storage into an established CentricStor archive solution is minimal--less than $200K--saving more than $500K on hardware without consideration for backup administration costs.

In closing, CentricStor as a component in ILM makes sense. The archiving process manages access to data, and the ability of CentricStor to provide a large, segmented data cache with separate migration policies enables CentricStor to perform as the secondary storage layer.

It is assumed that long-term data should go to CentricStor because cost-per-copy of data is approaching $0.0005/MB including management software today. CentricStor supports the non-disruptive addition of newer recording technologies and automated conversions, without impact to servers. CentricStor also has a feature for automatic media recycling to assure that all data is recoverable during the retention period. If CentricStor is in place for long-term data, then expanding its role to encompass secondary storage will eliminate $1M of hardware costs for ILM storage. Also, removing the backup requirement for the 13.4TB in secondary storage saves an additional $500,000. CentricStor saves about 20% of the total hardware infrastructure, simplifies administration, eases backups, and prepares the enterprise to adopt future technologies.

Classes of Data

                     Mission
                     Critical     Vital       Sensitive   Non-Critical

Primary Storage      Synchronous  Replicated  Replicated  Capacity
(0 to 30 Days)       Mirror Disk  Disk        Disk        Disk
Secondary Storage    Synchronous  Replicated  Replicated  Capacity
(30 to 90 Days)      Mirror Disk  Disk        Disk        Disk
Long-Term Retention  Nearline     Nearline    Nearline    Nearline
(>90 Days)

Classes of Data

Capacities with 5-yr   Critical     Vital       Sensitive   Non-Critical
Data Retention Policy  15%          20%         25%         40%

Primary Storage
0 to 30 Days           Synchronous  Replicated  Replicated  Capacity
(1.67% of total data)  Mirror Disk  Disk        Disk        Disk
Secondary Storage      Virtual Tape with
30 to 90 Days          Segmented Data Cache Controlled
(3.33% of total data)  By Separate Management Policies
Long-Term Retention    Virtual Tape
>90 Days               Managed
(95% of total data)    Nearline Archive

Mike Holland is vice president and general manager at CentricStor Business Unit, PeakData, Inc. (Niwot, CO)

www.centricstorusa.com