DAILY NEWS AND INFORMATION FOR THE GLOBAL GRID COMMUNITY / AUGUST 18, 2003; VOL. 2 NO. 33    ( Table of Contents )

Special Features:

GRID KNOCKS ON TELCOM DOOR

Just as different telecommunications networking technologies (e.g. frame relay, private line, asynchronous transfer mode, and virtual private networks) are used to build Intranets, Extranets, and public Internet services, various network technologies will be required to enable grids. To the extent that existing corporate and public networks can or cannot support the additional grid traffic and increased bandwidth demands, telecommunications suppliers will be tapped for new services and solutions.

Insight's research suggests that the emergence of grid computing technologies has broad implications for telecommunications carriers-ranging from internal implementation, to changing network traffic patterns, to value-added services resulting from computing as a utility.

Internal Business Solutions

Telecommunications firms are among the largest enterprise users of IT products and services. As such, the use of grid technologies could potentially improve efficiencies, lower costs, and improve productivity in the operations of their internal corporate data centers. Grid technologies could optimize use of their already-existing vast compute resources in research and development, network operations and management, and back-office IT infrastructures, solving critical problems and managing their own businesses more effectively.

In research and development, telecoms have many compute-intensive operations amenable to grid computing. These operations could be useful for applications that require massive engineering simulations, such as optimal design of a local mobile network. Keeping in mind telecom's constant drive to push network performance, network management and operations systems are mission- critical applications for grid technologies.

In the carriers' back-office, many potential candidates for compute grids or data grids exist. Long-running applications-such as payroll, billing, or customer data analysis-are just a few areas that could benefit from the improved efficiencies of these technologies.

Finally, as telecommunications firms, like many other technology firms, increasingly outsource elements of design, engineering, and production to partners around the world, the ability of grid technology to enable data sharing and collaboration among diverse and dispersed work units is of significant benefit.

Bandwidth and Traffic Patterns

Just as Intel has endorsed peer-to-peer technology as a potential demand- generator for their chips, so should telecommunications firms view grid and on-demand computing technologies as drivers of bandwidth demand, as more data, applications, and messages are exchanged across the network. This is completely consistent with the observation of most researchers in high- performance computing; when one makes resources more accessible to users, demand and traffic increase. It is logical to assume that when more resources are made available to nourish a given demand, more resources will be used.

Clearly, many of the scientific and research applications that will be running on grids over the next few years do have huge bandwidth requirements, typically in the 10-30 Gbit/s range. The NSF-funded TeraGrid project alone consumes multiple OC-192 metropolitan wavelengths and OC-768 long haul wavelengths connecting research facilities in Chicago and southern California. There are many other similar projects in play today in the US and around the world. The founder and CEO of one of the leading small grid software companies recently stated that in his experience, "Whenever a grid goes in, bandwidth goes way up."

While Insight's analysis suggests that the telecom network's role in the success of the overall grid structure is critical, organizations with both research and commercial capacities have put more focus on the computation aspect of grid computing than on the network (or grid) aspect. The grid community is just beginning to address the network as a critical component of the overall platform, and is just beginning to understand how networks can best serve grid computing environments.

To date, few (if any) public benchmarking studies have been performed regarding the impact of grid computing on local, metropolitan, or wide area networks, as confirmed by several leaders in the Globus Project. Although major bandwidth studies with respect to telecom demand and grid computing are occurring, these studies have been closely guarded.

Next-Generation Telcom Services

Although it is too early to speculate on the specific characteristics of local, metropolitan, and wide area network (WAN) services that would best meet the price, performance, and functionality requirements of grid computing applications in the future, grid community vendors and researchers have nevertheless expressed a variety of opinions on this topic.

• Fast Ethernet connectivity - Grid computing will likely drive demand for native Ethernet services, as there is increased need for processor-to- processor meshed connections. When one looks closely at a grid network architecture (such as the TeraGrid), it is clear that the connectivity among the many application servers, databases, computing resources, and users is most effectively accomplished by Ethernet. The pricing, however, will have to be palatable, and bandwidth-on-demand services will have to be available.
• Options for quality of service (QoS) and security - Beyond fast bandwidth, some grid applications will clearly require higher-level functions such as QoS and security. Characteristics like QoS are most important for real-time applications (i.e., online gaming and transactional processing) that may utilize grid technologies in the future. Many other potential applications will not require real-time processing, and thus will not have such stringent QoS security requirements.
• Latency - With good caching mechanisms in the application and middleware, the effect of latency-the time it takes to get information through a network-can be minimized. Additionally, the latency requirements will depend on the specific application that is grid-enabled. With loosely-coupled applications such as those in pharmaceutical companies, latency is not such an issue. With tightly-coupled applications such as those in aircraft design, latency becomes a bigger issue than bandwidth.

Potential Roles for Telecom Organizations

The first potential role that telecom organizations could play is in the production and delivery of computing as a utility. Drawing an analogy to players in other utility industries, several roles can be identified. At one level, there is the infrastructure play, where the telcos themselves could generate compute power from their own vast resources. Many of the telecom companies have created direct consumer delivery and marketing organizations and could surely leverage this expertise with their existing customer base.

Other organizations (traditional computer suppliers or telcos), could focus on the business-to-business marketplace by brokering the supply and demand aspects of compute power across different industries. Many large, multinational companies are intrigued by the prospect of generating revenues from their existing IT infrastructure, if these resources can be securely derived and delivered to a non-competitive company in need of such services. In the role of "computing brokers," telecommunications firms could match up customers with appropriate computing power/farm capacity.

As illustrated by grid pioneer Butterfly.net, grid technology can provide a "meta-operating system" upon which service providers can develop new collaborative applications, where vast compute power is distributed to diverse customers across the market spectrum-from consumers, to small and medium businesses, to global enterprise customers. By partnering with grid software providers and application software providers, telecom firms can re-architect applications to run efficiently over a grid infrastructure, and perhaps even incorporate enhanced service elements into the networks themselves. Such service elements might include various kinds of directories, security mechanisms, and the resource management mechanisms inherent in grid and on- demand computing.