Networking the intragrid

In a previous column I introduced the next “killer” application: the grid. For the grid to succeed in increasing corporate productivity and profitability, each evolutionary stage must be a success in its own right.

The first stage, the intragrid, will let different groups or organizations within a corporate facility share resources, minimizing capital expense. Two forms of intragrid will exist:

– Cluster or compute grid: This is the return of the data centre, or glass house. Using grid virtualization technology and Web services software, compute resources might not physically reside in one room but will logically exist as one compute resource to users and applications.

– Data grid: This is the ultimate manifestation of a storage-area network (SAN). Using grid and SAN virtualization technology, and Web services software, users and applications can access numerous database islands or silos of information simultaneously in a manner similar to a single local database.

To optimize intragrid performance, corporate network topology can be divided into three logical networks: a data centre network, SAN and backbone LAN interconnecting the two.

All upgrades to a data centre network must be optimized for the compute environment, accommodating legacy and new open interfaces such as Enterprise Systems Connection and Fibre Channel, in addition to possible vendor-proprietary channel interfaces. The data grid net also requires an optimized environment, most likely using interfaces such as InfiniBand and Gigabit Ethernet evolving to 10G Ethernet. However, no single interface standard might ever exist for the intragrid. Therefore, the only way to achieve connectivity at the optimal speeds required for grid applications will be to use optical wavelengths.

The grid will bring optical technology full bore into corporations. Wavelengths assigned to computing clusters or SANs are ideal for solving the interface, aggregation and bandwidth problems of intragrids. Immature products using service provider dense wavelength division multiplexing metropolitan optical equipment exist but are too expensive for most organizations. From a distance standpoint, the metropolitan-area network is neither a data centre nor a SAN. A better approach, using coarse wavelength division multiplexing and mesh optical switching or cross-connect technology, will create an optical LAN tailored to most organizations’ needs and budgets.

Architecturally, a PC LAN is not part of the grid except for user access. Therefore, it is not a candidate for upgrade as long as it allows as a minimum 100Mbps switched access to the desktop and Gigabit Ethernet access to the backbone using multicast and quality-of-service technology.

The last piece of the intragrid is the optical backbone that will connect data centres, SANs and PC LAN using wavelengths. This does not have to be an intelligent network component. In the optimal scenario, the backbone is formed using an optical switching cross-connect. All of the active operational grid intelligence is located in the connected distributed networks. Using this technique, a compute grid or data grid can be logically formed on demand under software application control.

Finally, add to the enterprise optical infrastructure new network-intelligent components for VPN configuration and management, application content routing/switching, Web services XML routing, grid load balancing and caching, and management software. This architecture will use a signaling layer built on Generalized Multi-protocol Label Switching, which controls end-to-end dynamic policy-, user- or application-based allocation of wavelengths.

Early adoption of intragrids has produced startling benefit estimates – up to a 25 per cent reduction in research and development costs coupled with a 25 per cent reduction in time to market for a new product. Grid technology will leave its mark beginning in 2004. But will the communications infrastructure components be there to meet customer intragrid implementation needs at the right prices, or will this be the impediment that limits corporate productivity growth?

Dzubeck is president of Communications Network Architects Inc., an industry analysis firm in Washington, D.C. He can be reached at