% fortune -ae paul murphy

Sizing, configuration, and the Unix reseller hurdle

One of the things that will continue to hurt sales of non x86 gear into small and medium business this year is the difficulty of sizing and configuring the gear. Now you'd think there should be nothing to that, but in reality you're just not going to get it done without spending a lot of time with sales people fronting for customer engineers - and that's if you're positioned to deal with them instead of reseller staff whose qualifications consist of being assigned to work the alphabetical block some reception thinks you belong in.

The big difference is this: with Unix there are many good choices for everything from hardware configuration to software; but with Wintel so much of the market belongs to one company, and the boxes are so interchangeable, that the effect of customer choice is negligible.

If, for example, you had to run the database backend for a 1,000 user organization on Wintel you'd be so likely to use SQL-Server that other choices need not appear on major vendor web sites - and you'll use pretty much the same hardware too whether you pick brand A or brand B, processor A or processor B, and this memory configuration or that. As a result companies like Dell and Compaq only really differentiate on advertising and can put detailed looking configurators on their public websites without exposing their customers to significant risk of configuration error.

With Unix, however, you face choices with real consequences - and using something like Sun's on-line store to configure your system is correspondingly high risk.

Thus Dell lets anyone do an on-line configuration for a a two way cluster of four socket, quad core, Xeons fronting a 6TB Sql-Server implementation - and will take the order at $224,166, because if the customer's out a couple of controllers, disks, or licenses, the absence of architectural choice means they'll smile while making the corrections involved.

But how do you get the comparable Unix configuration and number?

IBM's solution, as the remaining big commercial PPC supplier, is to force just about all transactions through an expensive (but effective) sales process based on personal relationships between sales staff and customers. From a cost perspective this is terrible, but as a consultant I've often recommended that clients pay simply for the assurance that comes from being able to trust that the customer engineer behind the sales quote really does have some clues about how the hardware and software interact in use.

Sun's solution has been to directly take on some big customers, tolerate a few technical resellers, tell most small customers that they're on their own, and fob the mid range off on resellers whose commitment extends all the way to this week's receivables.

As a result only deep dipped techies and those with either personal contacts in the company or decision making roles in very large customer sites have quick access to the expertise needed to get these decisions right - and the on-line store provides nearly infinite scope for the innocent to get them wrong.

You can, for example, use it to spec out that DB system mentioned above as a pair of T5440s fronting a 12TB J4200 disk store ($241,800); as dual T5240s fronting a 22TB 7210 storage server ($199,240?); or as a custom blade configuration containing two 6340s and 12TB of disk (about $180,000) - and that's before choosing between MySQL and PostGres - or Oracle, Sybase, or DB2.

Since, in practice, I don't think there's a real way to be sure about any complex configuration without testing your application on multiple alternate configurations, the questions most customers face are:

  1. how do you know what configurations to try? and,

  2. how do you arrange for the trial?

Sun offers test lab access in centers around the world and for bigger customers they'll help set up and do the test - but like the try and buy program (for which you don't get engineering support either) it'll usually take four months or more just to get there.

Worse, if you're not a big customer that test won't be free and not only could your cost of using the test center exceed the potential savings from switching to Sun - but you've also got to know that your reseller makes more money selling Wintel than SPARC and can be expected to use the time you spend waiting for Sun to quietly remind your bosses that saner people buy his Wintel line.

So if you're determined to go Unix, how do you deal with configuration risk? One good option is to go halfway: Linux on x86, because that reduces your choices, and therefore your risks, to software.

Another is to grit your teeth about the cost and buy from IBM - and a third is to learn enough about the SPARC/Solaris options to backstop your own sizing and configuration decisions by waiting out Sun's try and buy program.

But don't tell your bosses about these choices: because their easiest and most natural response to Sun's stratification of the market between the Connected and the Condemned (to IBM or Lintel) is to transfer the risk from configuration to operations by forcing that Wintel decision.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.