Technology: Utility Computing

Ashok K. Roy and Priya R. Roy

Utility computing will enable a university or a consortium to share hardware resources and draw on storage and other IT resources as required.

Dr. Ashok K. Roy is Director of Finance and Administration and Residential Programs and Services at Indiana University. He is the author of 46 professional publications. Priya A. Roy is a student at Bloomington High School South in Bloomington, Indiana.

Universities, like every other type of organization today, are assailed by changes in a dynamic environment.

Approximately 70 percent of university information technology (IT) budgets are spent on maintaining current systems. However, to gain an advantage over rivals, universities need internal operations and IT services that can adapt quickly to change. And it is here that utility computing comes in by breaking a paradigm—that one application runs on one server—by turning IT service provision into a utility service and providing resources where they are needed.

Utility computing will enable a university or a consortium of universities to share hardware resources and draw on storage and other IT resources as required. IT resources become more flexible and cost-sensitive to demand as the connection between hardware and the delivery of IT services is facilitated in a utility computing architecture. Further, consolidation of data center resources between or among universities could achieve a simplified model of their data resources.

Utility computing is essentially a systems integration of all components in a local integrated framework for greater functionality. In this concept, a shared IT infrastructure architecture is provided via demand-based services with pricing based on service usage and proven ongoing reductions in both fixed baseline and unit costs. Obviously, a number of elements have to be integrated into a utility service.

First among these elements is grid computing, where servers are networked together to perform as one large system. One well-known example of grid computing (a.k.a. clustered or distributed computing) that runs over the Internet is the ongoing SETI (Search for Extraterrestrial Intelligence) project in which thousands of users are sharing unused processor cycles and access to large amounts of data to help search for signs of “rational” signals from outer space.

Most servers utilize only 10 percent of their processor capacity. By networking the servers together and splitting workloads, the spare capacity in all available machines can be used to run applications, thereby increasing utilization. By adhering to the correct standards, different systems can be incorporated into a heterogeneous grid.

Currently, server farms, which are entire floors or buildings crammed with computer servers, tend to be for specific applications. Applications that are data-intensive (querying very large databases in real-time) or compute-intensive (requiring lots of computing power to do applications) can be run across, for example, 10 different server farms, each with 20 different servers. This can lead to cost reductions as fewer systems carry out the same number of tasks.

The debate over server farms sharpened during the severe energy problems in California. Server farms, unlike auto and microchip plants, have to operate 24 hours a day and maintain a constant temperature of about 68 degrees. But since servers don’t always run at full capacity, a server farm that has a 200-kilowatt capacity may use as little as 40 kilowatts per square foot of energy, making it on a par with an integrated petrochemical plant in terms of energy usage.

A second element of utility computing is that data and applications are shared between common systems. Computing resources are brought together to improve systems management and then accessed through wide area networks (WAN). Data is no longer bound to a single system and becomes “network resident” (meaning that data resides in the network consolidated data center). Consolidated data centers also permit the sharing of common resources, such as power units, storage, and disaster recovery. This leads to cost savings.

Blade servers (independent servers with their own processors, memory, storage, network controllers, operating systems, and applications) provide low-cost commodity systems that can be coupled through high-speed networks in the implementation of grid and utility computing. Blade servers simply slide into a bay in the chassis like books in a bookshelf (freeing up space and enabling higher density).

Universities can keep frequently accessed digital media files in their cache systems so that students can download the files without leaving the universities’ networks. Based on IBM blade servers, Napster permits universities such as Pennsylvania State University to save bandwidth while offering downloads. It is estimated that a university could save approximately $50,000 per year in bandwidth-related expenses alone.

Grid computing is widely used by investment firms (e.g., Charles Schwab), banks (e.g., JP Morgan Chase), and insurance companies (e.g., Prudential). The grid that connects university supercomputers at Indiana University and Purdue University surpasses the teraflop level (a trillion operations per second) of computation.

Carnegie Mellon University, University of Pittsburgh, University of Illinois, University of California at San Diego, California Institute of Technology, and others, in addition to Indiana University and Purdue University, are also part of the National Science Foundation’s TeraGrid. Europe’s answer to TeraGrid is called the DEISA. Harvard University uses IBM technology to create its Crimson Grid for collaborative research, while the University of Melbourne in Australia is in collaboration with Singapore Computer Systems.

Investments in IT, like other investments, are made to achieve an organizational goal, such as improving a business process or infrastructure, or addressing a new market. Most state-supported universities today face enormous pressures to control costs. In spite of using smaller systems and new network-centric structures, IT systems in most universities are inflexible and represent high fixed costs. While the cost of hardware is falling, reducing or removing fixed costs is a real help.

Legacy mainframe systems still manage core processes and are surrounded by client-server environments that provide services around the university and via the Internet. Expensive server systems typically operate 20 percent of the time, while storage devices are active 30 percent of the time.

The key lies in flexibility and adaptability in order to have the ability to reduce IT resources and control costs. In the event of growth, linking systems together with network access and demand-based capacity will provide universities the flexibility to buy in computing power as and when they need it. Utility computing offers the potential to control fixed costs and deliver business flexibility. By providing clustering and workload sharing, utility computing via grid computing with Internet protocol standard networks and consolidated computing centers enables smaller systems to be built up to provide the same resources that would require a much larger system.

Apart from the potential for reduced costs and flexibility, utility computing could help universities in integrating, analyzing, and reporting large volumes of data for compliance and control from regulatory and state agencies. In the future, it will be possible to purchase IT services in a demand-based model from external utility services

Utility computing will become an established business model in the near future. It is possible that even more radical technological changes may take place by virtue of sharing these utility capabilities across universities through standardized network protocol. For example, the aforementioned TeraGrid at Indiana University and Purdue University may give birth to major radical technological leaps in the life sciences, nanotechnology, genomics, global weather, and chemical catalysts.

In addition, utility computing could allow universities to reduce the number of servers by re-directing processes to spare resources and also to consolidate data centers. In time, it is possible that universities would be able to purchase IT services in a demand-based model from outside utility computing services and be able to manage IT usage and resources via service-level agreements.

Because of the use of utility computing models and grid computing technologies, the potential economies of scale benefits from a common infrastructure appear to be huge at this time. The question is whether, in spite of challenges, universities have the appetite to make these strategic choices. o