% fortune -ae paul murphy

Unix vs Wintel/DP: the cost/benefit issue

I've talked to middle managers who sincerely believe that a PC costs about $300 and a 2,400 user system should therefore cost about $720K - before volume discounts.

Similarly there are DP people who'll tell you that a single free Linux license at $40K per year will support thousands of instances on one IFL - serving 2,400 desktops entirely on freeware.

And, of course, everyone knows that a Unix server costs a couple of million bucks to put in - and then fails all the time despite the hundreds of thousands it eats in annual consulting and support fees.

Such views are delusory - in reality, fixed infrastructure costs for Unix and Windows favor Unix (no surprise given that several free Unix systems run on Wintel hardware), while getting vaguely comparable performance on zOS/IFL costs at least an order of magnitude more.

The most surprising thing about this, however, isn't the nonsense many managers believe but that the cost differences we consider important at the personal level and within IT, really don't matter much for larger organizations: what counts at the enterprise level are the operating costs, risks, and limitations imposed by whatever infrastructure is used - not the start-up cost of that infrastructure.

Thus the mistake that matters when people believe obvious nonsense on Wintel/DP costs has little to do with the specific errors they're making and nearly everything to do with the hidden assumption that systems architecture is an IT issue with no significant consequences for the organization - basically the problem is the belief that a computer is a computer is a computer, and therefore that the choices between them may entertain the IT troops, but are of little strategic relevance in the boardroom.

Imagine, for example, that we have a 2,400 user government organization in which:

  1. about 30 people in the financial review unit think they absolutely must have the very latest Microsoft Office desktops;

  2. about 20 people regularly use Adobe publishing and pre-print tools;

  3. about 240 need regular access to highly customized PeopleSoft financial and HR tools;

  4. about 1400 have some need for basic WP/Spreadsheet and presentation software;

  5. about 2,200 need daily access to a custom tracking and scheduling application;

  6. about 2,400 claim to need browser access and email;

  7. there are about 70 other applications, many of them locally grown spreadsheet, BASIC, or SQL-Server applications with the usual lack of documentation, backup, or auditability. On average each of these is thought to have about eight users, one of whom has largely rebuilt his or her job around servicing it;

  8. there are an estimated 15 to 25 data transfer and/or remote identification "applications" (many are scripts) that are uniformly said to be mission critical but are documented only with respect to their original forms - not as they are today; and,

  9. there are an unknown number of third party applications running on various servers really known only to those directly involved with them - and many of those people detest IT and will neither willingly report what they're doing nor participate in a change management program.

Right now no user manager, however senior, can make more than truly minor procedural or program changes without working through a formal IT assessment, technology change management, and (often) budget management process first.

The actual cost of the infrastructure used to meet these needs with wintel/DP is largely unknown - the book values are pure fiction. The outsourcer managing the desktop evergreen program values each desktop at about $1,400 inclusive of core desktop licensing and first level support; but exclusive of server infrastructure and custom applications, including Peoplesoft. There are about 160 physical servers in data center racks with recently purchased units averaging nearly $9,400 - much of it in licensing. There's an external shared data pool with its own servers - and nobody in the data center seems to know how many network controller racks and/or devices exist because these are administered by yet another third party. Bottom line? the senior IT people give $3,000 per user as a working estimate (about $7 million in total) and while that seems a little high for current replacement cost given government wide licensing for some Microsoft and other products, it's probably in the ballpark.

If I were to imagine replacing all this with a Solaris/SPARC system I'd probably think in terms of four identical units -largely because the organization is about evenly spread across contiguous floors in two nearly adjacent buildings and the idea of putting systems into opposing top and bottom corners in both buildings appeals to me. Each would have four T5440 servers with a 7410 (dual controllers, Flash, 40TB, 4 x x86, 128GB) data store in a rack with a UPS, highspeed router, and some 64GB, 4 processor, x86 Wintel server - and each would list at a bit under $600K for a net 20TB of fully mirrored data storage, 144 network channels, and 1024 virtual 1.4Ghz processors accessing 512GB of RAM.

Users would get 22" Sun Ray3 displays - at about $2.7 million in total including smart cards and software.

All the existing networking gear, including cabling, would go - to be replaced by an all optical system with no more than three devices in the path to the user: about $300K.

About 30 users would get their own little PC network - about $70K inclusive (about half in licensing). Another 20 would get Macs - about $80K inclusive (about two thirds in licensing). In both cases the "outside" backup processor would be in the nearest data center rack and double as the primary processor for other users needing occasional access to PC or Mac software.

The primary application was first developed using COBOL/IDMS in the 70s, converted first to a Vax running C/Forms in 1991, and then immediately to Adabas/Natural on the government's outsourced data center, is currently running as a windows client application accessing SQL-Server, and has been the subject of numerous successful redesign and redevelopment projects - none of which reached production. Today it looks almost trivial to do as a web application: about thirty data entry screens, perhaps 200 embedded functions, a dozen or so reports, and no known live interfaces to other applications.

In moving this to Solaris/Sun Ray I'd probably get two independent teams to convert it to PHP with MySQL - say $60K each - and then quietly start a re-examination of what the thing is really supposed to do and how it does it, preparatory to looking for a commercial or open source replacement (Although, of course, if I were doing this for real, I'd make porting it with guarantees and no nominal cost a condition of the deal with the Solaris/Sun Ray vendor.)

Each of the hidden applications would have to be found, documented, and evaluated during the conversion. In practice most are trivial and conversion, while possibly traumatic for those most closely dependent on control of them, fairly easy. Accommodations can, however be made if warranted - remember each rack will contain an x86 server able to provide Wintel support if and where needed.

Most of the other software is free and the Peoplesoft stuff is expected to transfer essentially unchanged despite the extensive customization effort ten years ago. As a result, and allowing about another quarter million for "surprise" software licensing and adaptation requirements, the whole mess should come in at a bit under six million - very roughly in the same ballpark they are now.

So where are the differences if they're not in capital cost? They're all in how using these two systems drives organizational behavior.

With Solaris/SPARC and Sun Ray:

Although these kinds of differences drive organizational value they're hard to measure but luckily there's a general rule that applies here: the simpler and more effective an engineering solution is, the lower its long term cost - and so when you go from the complexities of Wintel/DP's variation on the Rube Goldberg machine to the simple elegance of Unix with smart displays, you get measurable savings in IT operating expense.

Look at the whole thing on net, and the tangible savings from dropping client-server in favor of Unix with smart displays come mainly from staffing reductions, with some bonus monies coming from savings in annual software licensing and upgrades - but this is just the ginzu knives effect: the real value here is in the intangibles: in increased user productivity, in decreasing turn-over, in the elimination of most forms of software security risks, and in the near total elimination of system failures.

Thus the bottom line here is simple: the Unix system costs about the same as Wintel client-server to put in, but costs a lot less to run - and works significantly better on all the parameters users care about.


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.