Misaligned incentives drive complexity in IT. Cloud aligns incentives better.
By Juergen Urbanski
The crisis in IT is not about performance or features, it is about complexity. Complexity makes the enterprise choke on IT operations cost
Performance, in terms of CPU, memory and network bandwidth, is becoming an increasingly abundant commodity that simply gets better, cheaper and faster each year. The time has passed when any one company can create a sustainable competitive advantage based on hardware alone.
However, there are two scarce commodities that become increasingly prominent year after year relative to the declining costs of the above: latency and human attention. Latency is forever constrained by the speed of light and the distance between elements of a distributed system. IT labor does not ride on Moore’s Law either, to the contrary. As systems increase in scale, complexity increases and more human attention is consumed by the number, relationships and diversity of IT assets in the data center. Since the number and variety of nodes in an architecture that must be “supervised” by the administrator increases, there are more combinations of things that can go wrong.
Organizational models that date from an earlier, less complicated world contribute to the escalating cost and complexity.
Storage landscapes for instance have evolved as a result of customer efforts to minimize upfront cost and vendor efforts to maximize margin extraction. Paradoxically, a customer’s motivation to push back additional capex as long as possible may actually increase opex and lifetime TCO. Since customers are often locked in once the upfront investment is sunk, vendors can realize high margins for software maintenance contracts and other services and support.
Individual buying agents are almost never responsible for costs over the entire lifecycle of the project.
Many customers consider price per GB for storage a top criterion for new purchases, which tends to overlook operational complexity and inefficiencies, particularly as the data ages and gets moved onto secondary and tertiary storage. All too often, the person who negotiates the up-front price of the hardware and software licenses is not responsible for subsequent operating expenses.
Even the choice of block- versus object-based storage is partially influenced by organizational disconnects.
A major difference in usability occurs with the distinction between block-based (SAN’s, pools of LUN’s, volumes, etc.) and object-based (files, database) storage systems. Block-based approaches are about the management of pools of blocks which, until allocated and used, are fungible and substitutable. This perspective is appealing to administrators whose job it is to store data “centrally” (i.e., in some location convenient to them). Object-based approaches (files in particular) are more convenient to applications and users because they can leverage the consistency semantics of the file system, whereas the purely block oriented approach forces the introduction of the concept of a “single master”, along with its attendant complexity and problems. Block oriented storage, limited by the crude semantics of the SCSI interface, imposes the constraint of a single master to serialize the updates to the store, otherwise data could become inconsistent (i.e., different copies have different values) or worse still, corrupt (i.e., volumes become “un-mountable”). This leads to storage silos and devastating complexity when capacity limits are reached or things fail because applications are written to assume that they have exclusive control of the underlying stack.
Product vendors are not incented to reduce complexity on the data center floor, unlike service providers.
The extraction of margin by the vendors turns from the up-front capex costs into service and support fees to fix the bugs, inefficiencies and inadequacies in these products long after the sales transaction based on a “price per GB” criterion has taken place.
A side effect of this profit maximizing behavior is insufficient attention to customer total cost of ownership. For too long, vendors have been able to get away with products that are not designed, architected and tested with a view towards eliminating downstream operational complexity. Example pitfalls we see frequently include lengthy deployments, management that requires highly specialized skills, poor scalability, superfluous functionality and consumption of more storage space than necessary for specific functions.
Open source does not alleviate this conflict, because business models for open source are often founded solely on maintenance and services margin.
Crucially, a pay-as-you-go model that places the burden of complexity on the service providers seems to align incentives much better. In the long run, that may turn out to be the biggest benefit of cloud computing and storag
Tags: Cloud Computing, storage