Watch for the Grid

There's too much going on here to ignore. Folks from governments, research institutes, universities, and big corps are gathering together in sparely furnished conference halls, talking in excited, almost embarrassed tones, about something called "the Grid," with techie phrases like data mining, Web services, information utilities, and virtual organizations.

IBM Corp. (NYSE: IBM) is positioning itself to build a whole services business around it, while Hewlett-Packard Co. (NYSE: HPQ), Intel Corp. (Nasdaq: INTC), Microsoft Corp. (Nasdaq: MSFT), and Sun Microsystems Inc. (Nasdaq: SUNW) are all developing Grid strategies.

In homes around the U.S., kids are logging onto Butterfly.net and gaming in ways that depart radically from Microsoft's hopes for the Xbox. Protocols are being designed, virtual organizations envisioned. It sounds like the beginnings of the Internet years ago: a technology initially designed for the military or research organizations finding its way into the imaginations of capitalists, children, and a few of us who are not quite either.

This time around, it's about taking what there is on the Web and adding a method for tying together servers, storage, and computing resources to create virtual organizations, virtual worlds, or virtual supercomputers.

Consider how having affordable access to a supercomputer could help your business. You could offload computation-intensive projects onto the Grid (ASIC design, 3D modeling, scientific calculations, genome mapping, pharmacology, digital movie effects, you name it) and go about your day's work. Different organizations, different companies, or even different governments can use the Grid to collaborate, demonstrate, and create new projects (warfare and its devastating effects, predictably).It's got everyone excited, but just how entrepreneurs can participate today is not immediately clear. There are a lot of folks reading through arcane distributed processing literature out here in Silicon Valley, scratching their heads.

A good place to start watching the Grid is the Globus Project, initiated by Ian Foster and Carl Kesselman in 1999 as a blueprint for a new computing infrastructure to enable scaleable virtual organizations. As you'll see, Globus is an open-source toolkit for enabling computers and applications to be participants in a grid system.

Grids, according to the Globus Project, are "persistent environments that enable software applications to integrate instruments, displays, computational and information resources that are managed by diverse organizations in widespread locations."

Today, the focus is primarily on software, a middleware that provides the glue between users and distributed resources out in the network. That middleware needs its own set of protocols, its own method of ensuring security and distributed resource management. It will rely on the development of APIs (application program interfaces) and on a wicked fast, data-optimized network infrastructure. Globus is also developing applications that are purpose-built for grid computing and has released a couple of versions of its toolkit for developers.

Also take a look at the Global Grid Forum, which is going about developing the concepts of an Integrated Grid Architecture that will facilitate the development of grids, large and small, around the world.The whole world is involved in this, which leads one to wonder why so few telecom companies are making noise about it. Europe has devoted significant research dollars to grid computing, as has Japan, Canada, and most other developed nations (see Dot Hill Enters the Grid, Particle Physics Ups Storage Ante, and SAN Gear No Go for Grid Computing). Sounds like the Internet in 1995, doesn't it?

Ask Steve Mullaney, VP of marketing at Force10 Networks Inc., about the Grid and he lights up, looking 16 and about to get his first car. (This is good for a marketing guy.) Telcos are no place to make hay this year or next, and going after the enterprise runs you into Cisco Systems Inc. (Nasdaq: CSCO), hundreds of accounts, working through VARs, building lots of relationships... But the Grid: Hey, that's a few big players with big plans for services and the associated infrastructure. They've got the computers, the servers, the storage in-house, but they need switches, and fast ones. That's exciting.

Ahmar Abbas, managing director of Grid Technology Partners, has dived in headfirst after spending time at ONI Systems Inc. and UUNet. He's got a report out that makes a case for the Grid's lasting importance and considers it the "Next New New Thing." The Grid isn't just about IBM and supercomputers, but local grids as well, built inside a single company. If grid software can run on PCs, then all the desktops in a corporation can be part of a computing platform that can run processes that otherwise might be run on dedicated servers. In this sense, grid computing is a visible result of a trend toward peer-to-peer computing, harnessing the latent power of idle computers everywhere.

So what does this mean for the optical and data networking community? Two things: 10-Gigabit Ethernet and the reemergence of the ASP (application service provider). Here are the details.

First, 10-Gigabit Ethernet. Looking at a Grid networking architecture (such as the one here: www.teragrid.org/img/teragrid.sm.jpg), it becomes quite clear that connectivity among the many application servers, databases, and computing resources and users is most effectively accomplished by Ethernet. Gigabit Ethernet is clearly the most cost-effective interface for servers, storage systems, computers, and supercomputers the muscle of the Grid.Thus, it follows that if you want to hook all these together at the highest speeds possible you want little more than aggregated, clean throughput – none of the fancy packet processing that gets built into router ports, but straight-ahead forwarding. 10GigE switches, located at the boundaries of Grid nodes, will be on-ramps to the connective tissue of the Grid, the optical transmission networks. In the metro, optical Ethernet solutions can provide low-cost transport over dark fiber, via CWDM or DWDM, whichever makes the most economic sense for the provider.

Further down the road, Grid service providers will be asking for more than just forwarding performance out of their switches. Already, a great deal of work is underway to support security in distributed computing systems. Switches may have to incorporate higher-layer functions of security, QOS, etc., to remain relevant to increasing service demands from these new service providers.

Which brings us to number two, the reemergence of the ASP, or perhaps GSP, for Grid Service Provider. These GSPs will be quite distinct from our telcos – which may find themselves reduced to being little more than pipe suppliers – focused entirely on providing a robust Grid for customers to access for any number of computation- and storage-intensive applications.

According to Force10's Mullaney, online gaming is an early application of grid computing. "The infrastructure demands are huge to be able to support a scaleable gaming platform. That's perfect for folks like IBM to host people like Butterfly.net. It lets them focus on the gaming applications while IBM focuses on the infrastructure. It also lets Butterfly.net become profitable because they don't need to worry about how many servers to buy. IBM builds a grid infrastructure to support them and dynamically adds more compute resources as needed. In this way Butterfly does not have to go buy and install the maximum number of servers to handle peak loads."

Abbas of Grid Technology Partners sees things developing this way: "Grid Computing will gain a foothold in the product life-cycle/R&D side of the firms first. These groups already have defined computational requirements, as evidenced by their existing investments in high-performance computing and clusters. For example: electronic design automation applications at semiconductor companies; computational fluid dynamics applications at automotive and aerospace industries; bioinformatics and proteomics applications at big and small pharmaceutical and biotechnology firms. Most of these applications already have been parallelized for deployments on clusters and high performance computers. This phase has started already."The second wave of deployment will start when IBM, BEA Systems Inc. [Nasdaq: BEAS], PeopleSoft Inc., etc., have Grid-enabled their business-process-related applications. This phase, in my opinion, won't start until 2004/5. But the interesting part here is that once this does start, most companies would already have grids deployed for R&D applications – hence the deployment should be friction-free and swift. The CIO will be the decision maker here. The third phase will be Internet applications, including gaming."

What stands out here is how unique this opportunity is in today's market of ruined service providers. Telcos thought the Internet would justify their expansions, but the Internet has no real owner, so access was about all any carrier could provide, and access or transport gets commoditized quite easily – so before they knew it, telcos were charging less than their own costs just to get customers on their networks. It's a downward spiral in which telcos are still caught. Their answer is to add value to their commodity services, but that value remains locked in the way bits are transported.

"The main market for telecom equipment is today primarily from public-sector grid projects such as the Teragrid, etc.," says Abbas, a veteran of the telecom world. "The telecom component (including dark fiber acquisition) is roughly $200 to $300 million this year. While winning a portion of this may not have a material impact on, e.g., Cisco's revenue in percentage terms, it has worked out in the favor of Ciena Corp. [Nasdaq: CIEN], Force 10, Qwest Communications International Inc. [NYSE: Q], and Level 3 Communications Inc. [Nasdaq: LVLT] recently. However, just as Nortel Networks Corp. [NYSE/Toronto: NT] garnered substantial revenue selling their metro optical gear through the storage area application, vendors that gain early mindshare will be able to sell their wares around this application as enterprise grids start getting built.

"Secondly, Teragrid and its counterparts in Asia and Europe could be laying the foundation for a next-generation network, Today, Teragrid has 4 nodes National Center for Supercomputing Applications, the Argonne National Laboratory, the San Diego Supercomputer Center, and the California Institute of Technology (Caltech)