Economies of Scale have been much discussed in The Agile Executive since the recent OpsCamp in Austin, TX. The significant savings on system administration costs in very large data centers have been called out as a major advantage of Internet-scale Clouds. Unlike various short-lived advantages, the benefits to the Cloud operator, and to the Cloud user when the savings are passed on to him/her, are sustainable.
In this guest post, colleague and friend Annie Shum analyzes the various sources of waste in operations in traditional data centers. Like an Agilist with Lean inclinations who confronts an inefficient Waterfall process, Annie explains how economies of scale apply to the various kinds of waste that are prevalent in today’s small and medium data centers. Furthermore, she connects the dots that lead toward a Green IT option.
Here is Annie:
Harnessing Economies of Scale in Cloud Computing to Realize a Greener Computing Option
Scale Matters: “Over time, however, competitive advantage within categories shifts inexorably toward volume operations architecture.” – Geoffrey Moore, “Dealing with Darwin”
It is a truism that today’s datacenters are systemically inefficient. This is not intended as an indictment of all conventional datacenters. Nor does it imply that today’s datacenters cannot be made more efficient (incrementally) through right sizing and other initiatives, notably consolidation by deploying virtualization technologies and governance by enforcing energy conservation/recycling policies. There are a myriad of inefficiencies, however, that are prevalent in datacenters today.
Many industry observers lament the “staggering complexity” that permeates on-premises datacenters. Over time, most, if not all, enterprise IT datacenters have become amalgamations of disparate heterogeneous resources. Generally, they can be described as incohesive, perhaps even haphazard, accumulations. The datacenter components and configurations often reflect the intersections of organizational politics (LOB reporting structures leading to highly customized/organizational asset acquisitions and configurations), business needs of the moment (shifting corporate strategies and changing business imperatives to gain competitive edge or meet regulatory compliances) and technology limitations (commercial tools available in the marketplace). It should come as no surprise that human interactions and errors are considered a major contributor to the inefficiencies of datacenters: IBM reported that human errors account for seventy percent of the datacenter problems.
The challenge of maximizing energy efficiency begins fundamentally with the historical capital-intensive ownership model for computing assets to enable each organization to operate its own datacenter and to provide “24×7 availability” to its own users. The enterprise IT staff has been required to support unpredictable future growth, accommodate situational demands and unscheduled but deadline-critical events, meet performance levels within SLAs and comply with regulatory and auditing requirements. Hence, datacenters generally are over-configured and over-provisioned. In addition to highly skewed under-utilization of distributed platform servers, ninety percent of corporate datacenters have excess cooling capacity. Worst of all, according to IBM, about seventy-two percent of cooling bypassed the computing equipment entirely. Further compounding these problems for a typical enterprise datacenter, is the lack of transparency and the inability to control energy consumption properly due to inadequate and often inaccurate instrumentation to quantify energy consumption and waste due to energy lost.
The economics of Cloud Computing can offer a compelling option for more efficient IT: by lowering power consumption for individual organizations and by improving the efficiency of a large number of discrete datacenters. Although the electricity consumption of Cloud Computing is projected to be one to two percent of today’s global electricity use, Cloud service providers can still cultivate sustainable Green I.T. effectively at lower costs by leveraging state-of-the-art super energy efficient massive datacenters, proximity to power generation thereby reducing transmission costs and, above all, harnessing enormous economies of scale. To better understand how Cloud Computing can offer greener computing in the Cloud and how will it help moderate power consumption by datacenters and rein in run-away costs, a good starting place is James Hamilton’s September 2008 study on “Internet-Scale Service Efficiency” as summarized in the table below.
Resource Cost in Medium DC
Cost in Very Large DC
Ratio Network $95 / Mbps / month $13 / Mbps / month 7.1x Storage $2.20 / GB / month $0.40 / GB / month 5.7x Administration ≈140 servers/admin >1000 servers/admin 7.1x Table 1: Internet-Scale Service Efficiency [Source: James Hamilton]
This study concludes that hosted services by Cloud providers with super large datacenters (at least tens of thousands of servers) can achieve enormous economies of scale of five to seven times over smaller scale (thousands of servers) medium deployments. The significant cost savings is driven primarily by scale. Other key factors include location (low cost real estate and electricity rate, abundant water supply and readily available fiber-optic connectivity), proximity to electricity and power generators, load diversity, and virtualization technologies.
Will this mark the beginning of the end for traditional on-premises datacenters? Can enterprise IT continue to justify new business cases for expanding today’s non-renewable energy powered datacenters? According to the McKinsey article, the costs to launch a large enterprise datacenter have risen sharply from $150M to over $500M over the past five years. The facility operating costs are also increasing at about twenty percent per year. How long will the status quo last for enterprise IT considering the recent trend of Cloud service providers? Major players such as Google, Microsoft as well as the U.S. government itself have invested in or are planning ultra energy-efficient mega-size datacenters (also known as “container hotels”) with massive commoditized containerization and proximity both to power source and less expensive power rates. Bottom line: will the tide turn if the economics (radical cost savings) due to enormous economies of scale become too significant to ignore?
Despite the potential for significant cost savings, it is premature to declare the demise of traditional IT or the end of enterprise datacenters. After all, the rationale for today’s enterprise IT extends well beyond simplistic bottom-line economics – at least for now. To most industry observers, enterprise datacenters are unlikely to disappear although the traditional roles of enterprise IT will be changing. A likely scenario may involve redistributing IT personnel from operating low-level system operational tasks to addressing higher-level functions involving governance, energy management, security and business processes. Such change not only would become more apparent but will likely be precipitated by the rise of hybrid Clouds and the growing interconnection linking SOA, BPM and social computing. Another likely scenario is the rise of the mega datacenters or “container hotels” for Cloud Utility Computing providers. Although the global economic outlook will undoubtedly play a key role in shaping the development plans/timelines of the mega datacenters, they are here to stay. Case in point: by 2012, Intel estimates that it will design and ship about a quarter of the server chips (it sells) to such mega-data centers.
