The Holy Grail of Storage Efficiency, Part 4 -- Rethinking Cloud Storage Management
In both the MSP and the SSP models for storage clouds, it is the management that delivers the value, not the infrastructure.
In previous parts of this series, we looked at storage economics, architectural models for building storage infrastructure, the merits and demerits of enhanced controller designs and the role of virtual controllers, and where to host value-added functionality most effectively. All of these issues need to be carefully considered when planning an efficient storage practice. What remains to be discussed is storage management. We’ll tackle the topic in the final two parts of this series.
Our discussion follows from a thread from the previous column, which prompted a few e-mails from readers -- mainly vendors of storage clouds. Critics argued that the piece was too dismissive of clouds, which (according to one writer) solve the problem of comprehensive storage management that is proving so elusive both to the industry and to consumers.
One said that my treatment of "cloud storage" conflated services aimed at providing storage as a service with those providing storage in connection with some other objective, such as remote backup or analytics and management. The latter class of cloud vendors does not seek to sell storage capacity per se; they are selling a service that happens to include or relates to storage management tasks such as performance management, capacity management, or data protection management.
This delineation of cloud services is reminiscent of application and storage service provider-speak of the late 1990s. Some providers sought to distinguish their services, which consisted of remote storage management, from others who were providing actual storage infrastructure on an outsourcing basis, thus creating a distinction between MSPs (management only) and SSPs (managed infrastructure). The MSPs were providing externally hosted software with analysis and support to spot problems in the customer's infrastructure and to recommend fixes and workarounds. SSPs wanted to sell more capacity.
The distinction may be a valid one even in the context of today's clouds. Certainly, cloud services such as Fujifilm USA's Tape Library Analysis Service are an example of the MSP model. This service involves deploying a Read/Verify Appliance from Crossroads Systems at the customer site, using it to collect data on tape library usage that is forwarded to the service provider for analysis and feedback to the customer. The goal is to optimize tape performance by providing a set of technologies and skills that supplement on-staff capabilities with respect to tape management expertise. The result is typically a significant reduction in operational costs, demonstrable improvements in tape resource efficiency, and better data protection without increasing staff.
Conversely, a modern day SSP cloud -- think Amazon's S3 -- comprises an effort to sell "managed storage capacity." It is an infrastructure outsourcing effort that only works if the vendor can persuasively argue that the service delivers the same or better service levels in a secure way at the same or lower cost that consumers could realize from operating their own storage infrastructure on premise. There are basically three flaws in most cloud storage offerings today that engender doubt in consumers who look at this strategy.
First, cloud outages have occurred from time to time in both large and small SSP cloud providers that have received broad coverage in the trade press. Although providers are quick to point out that storage outages also occur inside corporate infrastructures, "our service tends to fail a bit less often than your infrastructure fails" is not a compelling marketing message -- especially not to companies that hold service providers to a higher standard in terms of SLAs than they do their local IT shop. Cloud SSPs try to argue that they are delivering a "better managed" storage infrastructure than most companies have today -- which may be true -- but they can't have it both ways. "Better managed" is supposed to translate to more available, reliable, and dependable, so the marketing pitch loses its luster -- for a time, at least -- whenever a publicized outage occurs.
Second, data assets hold a special place in the minds of even non-technical business management. They are not anonymous sets of 1s and 0s, but e-mails, contracts, customer lists, 401K details, etc. They are real information assets with real value either in and of themselves or by virtue of their costs to the company if illegally disclosed or inadvertently deleted. Again, the cloud SSP is held to a higher standard as the guarantor of security for data that is placed in its charge. It is true that on premise IT operators may lose data or that data may become exposed deliberately (via whistleblowers) or inadvertently (via hackers), but the cloud SSP is rightly or wrongly expected to be the ultimate data lockbox. Their security is expected to be air-tight and their data protection processes on par with those of the national security organizations. Recent highly-publicized security breaches involving clouds call this premise into question.
Third, as a practical matter, clouds are supposed to be cheaper than roll-your-own storage infrastructure -- or at least priced competitively when total cost of ownership factors (labor expense especially) are considered. Vendors tout prices like 17 cents per GB for storage, but neglect to mention the additional costs for wide area networks, add-on services such as replication and backup, and (in some cases) the cost to retrieve data once stored. They present papers arguing that cost savings in storage administrator labor, utility power, and facilities. more than justify their competitive pricing claims -- and they are correct in this assertion in many cases. However, sustainability of these claims over time is questionable.
From a technology perspective, disk continues to become less expensive as capacities increase. Within the next couple of years, as a function of new technologies such as bit-patterned media, it is reasonable to anticipate the appearance of 2.5 inch drives with upwards of 40TB of capacity. Within a year of release, these components will be available for less than .0005 cents per GB, offering sufficient capacity to store a smaller firm's entire data complement on only one or two units with a price tag of a few hundred dollars. The number of drives in play will be greatly reduced, yielding economies both in administrative labor and utility power. Moreover, improvements in tape capacities and falling prices of Flash drives may well see a new storage infrastructure model in which disk is less important than it has been in the past.
The bottom line: in both the MSP and the SSP models for storage clouds, it is the management that delivers the value, not the infrastructure. This provides a suitable segue way to the topic that needs to be addressed in this column: storage management. We’ll conclude our series by tackling that subject next week.
Your comments are welcome: firstname.lastname@example.org.