Speculation About Compaq

In May 1997, Tandem demonstrated a 2TB database scaling evenly across a 16-node Windows NT cluster. Although the database was spread across 700 disks, Tandem's NonStop SQL engine was able to perform parallel queries on the multibillion-row database, returning the results in seconds. NonStop Software for NT is 3 to 5 years ahead of any other NT database and clustering solution available. Now that Compaq owns this technology, the company has a huge and unique opportunity to influence the future of the NT market for database and clustering technology.

So why haven't we heard about Tandem's NonStop Software for NT since Compaq bought it? Compaq is a successful partner to software companies because its products are database neutral and operating system (OS) neutral. Compaq wants IS managers to think of its hardware whether they run SQL Server, Oracle, Sybase, or DB2 on NT, NetWare, or SCO UNIX. If Compaq favored a database, say NonStop SQL, over another database, say Oracle, Oracle could revoke Compaq's premier partner status. Compaq could not replace lost revenue from Oracle with revenue from NonStop SQL on Compaq. As a result of this dilemma, Tandem's advanced NonStop Software technology sits on the shelf.

What if Compaq dealt with this dilemma by selling Tandem's NonStop Software to a database vendor (say Oracle) that has clustering technology? This approach would not produce a good fit. Oracle's scalability philosophy is to build large symmetric multiprocessing (SMP) systems and to use shared-disk clustering. To use NonStop Software's shared-nothing cluster technology, Oracle would have to rewrite its Parallel Server on NT from the ground up.

What if Compaq sold NonStop Software for NT to Microsoft, another database vendor that has clustering technology? As my article, "The Future of Clustering," page 116 shows, Microsoft's direction is shared-nothing clustering, the same model that NonStop Software uses. In fact, Microsoft plans to provide NonStop-type functionality over the next 5 years. If Microsoft handed NonStop Software technology to its SQL Server engineers and told them to incorporate it now, the evolution of SQL Server could take a leap equivalent to skipping three upgrades. Microsoft could land so far in the lead that no other database vendor could keep up.

On the hardware high end, Tandem's Himalaya systems, which include a proprietary version of NonStop Software, are the leaders for large-scale highly available transaction systems. Tandem servers manage an estimated 80 percent of the world's automated teller machine transactions. What if Microsoft suddenly owned NonStop technology and could deliver it on high-volume, low-cost servers? You would be able to use 4-CPU servers as your cluster nodes and run some of the largest database applications in the world, achieving linear scalability. Forget 8-CPU or 16-CPU systems; you wouldn't need them. If your application ran out of steam, you'd simply add another 4-CPU node, re-partition the data evenly across all the nodes, and increase performance linearly.

What about non-database clustering? Microsoft would benefit greatly from purchasing this technology because Tandem's NonStop Software contains NT-based cluster software that already scales to 16 nodes (in contrast to Microsoft Cluster Server's­MSCS's­two nodes). NonStop Software works with Tandem's high-speed cluster interconnect, ServerNet, and has already demonstrated availability and scalability.

Selling Tandem's NonStop Software to Microsoft would be a win-win solution. Compaq would remain database neutral and an excellent partner to all database vendors. Microsoft would make a giant leap toward its stated direction in database and cluster technology.

In addition to the NonStop Software dilemma, Compaq inherited a similar problem with Digital Equipment's 64-bit UNIX. By owning a version of UNIX, Compaq loses its neutrality with server OSs. Will Compaq promote Digital's UNIX over SCO UNIX or NT? That strategy would result in a fundamental shift in Compaq's partnerships. More likely, Compaq will sell Digital's UNIX to Compaq's UNIX partner, SCO.

Two things need to happen to make this sale possible. First, Sequent needs to keep its agreement with Digital to help port Digital's 64-bit UNIX to Intel's IA-64 (Merced) chip. Second, SCO needs to provide a migration path to the new 64-bit version. If both things happen, Compaq can offer 64-bit SCO OS on its future IA-64 based systems, making SCO the leading hardware-neutral commercial version of UNIX. This win-win solution lets Compaq remain OS neutral and gives SCO access to a sophisticated 64-bit version of UNIX.

Compaq's Choice
Compaq has acquired incredible technology that it can shelve or use to change the course of the computing industry. The ball is in Compaq's court.