Network Computingwww.networkcomputing.com

Unfortunately, Oracle's new 10.7 NCA (Network Computing Architecture) version has some caveats. First, it was announced as shipping in January, so it's untested in real implementations. Oracle has created middleware, something PeopleSoft decided not to do (TUXEDO scalability has been proven in long-term customer implementations of other applications). Moreover, the Java-centric client approach may be held back while corporations deploy compatible Web browsers and other infrastructure. Oracle specifically requires very recent JDK/JVM (Java Development Kit/Java Virtual Machine) support--for local persistence, applet signing and so forth--and plans to use Sun and JavaSoft Activator technology to download its own JVM code to each browser. The Oracle application viewer applet itself is only 550 KB, but the JVM will be as much as 10 MB; then again, these change only a few times a year. Still, this may force organizations to focus more closely on Web-centric JIT (just-in-time) software delivery options (branch-office Web servers with applet replication, cascaded server or multicast-based push approaches).

There is also still the issue of downloading the screen formats for the application modules themselves. Organizations have noted that the two-tier Oracle Applications client module delivery was a key problem in deployment. While the new, small applet is much different, downloading screen description data every morning or as changes are made and modules are loaded is still an issue. Oracle is working to do more intelligent caching of both screen and application table data in the future.

Finally, Oracle has yet to publish significantly detailed network performance data. With preliminary data, it's suggesting that the new version will need 3 to 5 Kbps per user. Oracle noted that it's also tried to get down to a single round-trip for each field validation or final screen transaction commit function. So the new application should function better over higher-latency WAN links like frame relay. Still, it doesn't look like it will be as efficient as SAP R/3 or PeopleSoft 7.x. I expect it will take customers the rest of this year to get used to the idea of a Java-only 32-bit GUI environment and then take some more time to deploy the solution.

Back to the Future--3270 Reincarnated? SAP, in contrast to PeopleSoft, has a white paper that provides a formula for calculating users per link given a particular bandwidth while factoring in user think time between transactions. Of course, such formulas should be used only to gauge general loading; real loading can be influenced by many factors (particularly the variability of latency, types of transactions and the presence of other interactive or bursty file transfer traffic running over the same link).

But at least this is a good cut at capacity planning in this new world where everything is so much more complex than the old SNA 3270 days. Not surprisingly, applications like SAP R/3 do function practically like block-mode 3270, except now we have all the GUI bells and whistles to which we've grown accustomed.

The real networking problem with three-tier applications is no longer primarily the way each behaves in particular. In fact, as all the major ERP (enterprise resource planning) products move to three-tier, their networking details are less critical. ERP is becoming more like client/server messaging, where the basic paradigm is one where there are no glaringly disproportionate networking impacts. While there is some variability, more products will act similarly in a networking environment.

No, the latest problem with applications is the fact that everything runs on TCP/IP. Everything is on the same network, all competing with each other. As networking IT struggles to assess the option of quality-of-service offerings for "better than best effort" networking, identifying critical applications and maximizing their built-in potential (rather than fixing their faults) will become increasingly important (see "Traffic Shaping: Assuring Application Performance," at www.NetworkComputing.com/822/822colrobertson.html).

IT Can Save on Band-Aids Overall, we're seeing enterprise packaged application vendors move to three-tier architectures for their flagship applications. They are making traditional two-tier Band-Aids like Citrix WinFrame remote control less valuable to corporations. I still strongly believe that a well-architected application (three-tier, message-oriented middleware, and so on) can easily outpace any Band-Aid approach. Application developers still should be looking for ways to enable enterprise-scale networked deployments.

In fact, the new frontier for ERP applications will be enabling or "federating" separate business-unit (and geographic) instances of large-scale deployments. While this situation is often forced upon an organization because of politics (each unit or territory wants to own its IT solutions), sometimes the distributed operations would be more efficient from the networking perspective. Today, application broker middleware, for example, strives to solve this problem for disparate applications. Within a particular application system, however, vendors haven't taken enough time to design distributed instance scalability.

Although these changes will take some time to affect larger implementations, I see fewer nightmare application deployments in the future. And we can all use the sleep, can't we?

Bruce Robertson is a program director with the META Group's Global Networking Strategies service. He can be reached at Bruce.Robertson@metagroup.com.