July 23, 2003

Unix architecture yields cost, competitive advantage

- by Paul Murphy -
Last week I introduced the Unix business architecture and contrasted
its technology -- big servers with smart displays -- with that of the
Microsoft client/server environment. Today I'll look at the direct costs of both
architectures in the context of a typical midsized business.

The Unix architecture turns out to be cheaper, but more importantly it is also considerably more powerful as a tool for building and holding an overall corporate competitive advantage. That benefit comes from the architecture -- the combination of technology with management methods appropriate to the Unix/smart display environment -- and not just from the technology.

The question to ask about the Unix business architecture isn't whether it's cheaper -- that's obvious -- but whether it's smarter -- a much more subtle and context driven question.

Suppose, for example, that a company with hundreds of servers and several thousand desktop and laptop PCs
running Microsoft Windows operating systems and application suites is taken over
by a Ferengi. He promptly replaces every piece of licensed software in this client/server architecture with the nearest open source equivalent. He shuts down some servers but makes no other hardware changes and only minimal networking and access control changes.

Afterward the company finds it is no longer paying licensing costs. Applications seem to run faster because Linux is more hardware-efficient than Windows. Failure rates at the both OS and application level are down. Virus problems virtually disappear as a daily issue.

The new setup saves money, but is the company better off?

Some applications are better, some are worse (no one's complaining, the CIO is a Klingon) but
the company is still doing the same things in fundamentally the same way, because client/server is client/server
whether implemented with Wintel or Lintel. Users still face a cantankerous desktop PC, networking is
complicated, there are far too many servers, and administration is a pain. The change saves some money, but fundamentally does nothing in terms of affecting the organization's competitive position. A dollar saved is a dollar available
for something else, but is this enough to justify the hassle? In the real world, probably not.

To see beyond direct cost issues to competitive advantage, you need business context -- history,
structure, markets, and technologies -- to show what happens when you try to fit a systems architecture to business strategy in order to gain true corporate competitive advantage.

Consider the tens of thousands of small
to medium-sized manufacturing companies that parlayed some form of
early competitive advantage into strong growth during
the '80s and '90s, but now face intense competitive pressure.
In many cases these companies made their last conscious effort at strategic systems planning
more than a decade ago. During the '90s economic good times allowed senior management
to abdicate their responsibility for IT governance in order to sustain
a fragile peace between PC proponents and the older generation of IT
professionals responsible for key applications in Finance or Engineering.
Today, however, the original core applications are so outdated
that senior management is being forced to confront systems issues
whether it wants to or not.

Typically such a company has three choices:

  1. Upgrade the existing critical applications, often through an outsourcing deal, while formalizing
    the pre-existing drift toward an all-Microsoft client/server architecture outside the server environment housing those key applications.

    In many cases this is an extremely attractive approach because it continues existing practices while offering a non-confrontational way to fire the old time IT people running the existing applications and using the outsourcer's banking services to avoid recognizing short-term costs.

  2. Initiate an intentional strategic commitment to the Microsoft architecture, load up a Microsoft
    business application
    , bite the bullet on removing outdated infrastructure and personnel, learn how to manage a Windows-based system, and try to gain long-term control of the investment; or,
  3. Adopt the Unix business infrastructure, making whatever changes in hardware, software, and personnel are needed to succeed with it.

Of these, the first is usually a do-nothing strategy that merely postpones the decision between the other
two, so we'll deal with only those here. Here's what they look like:

  Microsoft client/server Unix business architecture
Desktops Fully "locked down" (no user-modifiable setup parameters or software) Windows XP PCs, 3GHz P4, 256MB, 17" Mostly Sunray2 17" smart displays; some 21" displays; dual Xeon workstations running Red Hat in
Laptops Fully "locked down" Windows XP laptops, 1.6GHz P4, 256MB, 15.4" Mac G4, mostly 17"; some 12"
Servers 42, in rack mounts, mostly dual Intel Xeon-based hardware, Microsoft Windows 2003 Server, SANs 3, Sun V1280s; 12 CPUs, 96GB, 2TB integrated Fibre Channel (2GB/sec) arrays
General use software Microsoft servers for email, scheduling, office file management, databases,
web services, etc.
All open source
Major application software Microsoft business applications (ERP/SCM), mainly Great Plains with multiple third-party and custom elements Full Oracle, Peoplesoft, or steel-industry-specific suite
Specialty software AutoCAD, some minor third-party applications LinuxCAD, multiple third-party and custom applications; some AutoCAD on legacy Wintel gear for customer conversions
Networking Switch-based internally, Microsoft NAT/proxy and firewall servers Integrated voice/video/data; all router-based with integrated VPN; Apple AirPort Extreme for job sites
Management focus Virus and related defense, keeping up with upgrades, hardware/software maintenance, battling failures, satisfying user change demand Helping users apply the technology in revenue generation
Implications for at-home computing Remote access requires Microsoft Windows-only VPN client No technical impediments to Windows or Unix; company management actively supports Linux/BSD at home

Buldane Steel is a fictionalized composite of several real companies I've worked with. The company
started out as a civil engineering partnership but now also makes and sells custom structural steel.
In the usual sales cycle Buldane tries to have its engineers engage with customers to develop
custom steel specifications before bid requests are issued, delivers directly to the job site, and provides project engineering services on site during construction.

The actual steel plant employs about 430 people. Another 180 work in the main office, about 75 of them in customer engineering roles and 33 in IT. There are 46 people in two sales offices, and 18 at a remote long term project office. The remaining 126 people nominally work out of the head office but are almost always on client sites.

The company is considering two alternatives for updating its systems.

The Microsoft client/server proposal

With this approach, the company would hire consultants to help implement a Microsoft Great Plains
ERP/SCM solution for the company, and upgrade servers and desktops throughout the company to HP hardware running Windows XP Professional and Windows Server 2003.

Component Configuration Quantity Unit cost1 Five-year total cost2
Engineering workstations HP xw6000; Dual 2.8GHz Xeons, 512MB, 36GB disk, 21" CRT, Windows XP Professional and Office XP 140 $4,680 $1,711,920
AutoCAD upgrades From V2 to Autodesk 2004 140 $1,795 $628,250
Office PCs HP d530 Small form factor; 17" flat monitor, 256MB, 3.0GHz P4, Windows XP Professional; Office XP 210 $1,912 $1,114,680
Notebooks HP NX7000; 1.6Ghz P4, 15.4"; 256MB Windows XP Professional and Office XP; basic carrying case, power pack, Ethernet adapter; HP wireless adapter 250 $2,048 $1,412,000
Servers HP ProLiant ML330; Windows Server 2003; 55 CALS; 2GB, 2 x 2.8GHz Xeons, 6 x 73GB Raid 0 disk;
battery-backed controllers
42 $33,137 $4,281,585
Microsoft business applications Mainly Great Plains ERP (including Financials, Reporting, HR);
some elements from Axapta, others custom developed by consultants
Includes 220 days on-site consulting support; total cost is
an estimate only. Annual support is 20%.
$725,000 $1,305,000
IT staffing Selected from existing staff 21 PC support, 3 management; 15% benefits $1,345,500 $6,727,500
Totals       $15,875,935
1Pricing generally from Microsoft, HP, and Autodesk Web sites as of July 1st, 2003

2The five-year cost estimates include 15% maintenance per year plus one PC hardware
refresh at 36 months and PC software refreshes at 24 and 48 months.

The estimated five-year cost for this alternative is $15.9 million dollars.

Notice, however, that we are ignoring a lot of costs. For example, printing and networking
costs are commonly excluded from proposals like this on the grounds that these don't change with acceptance or rejection of the proposal. In reality that argument assumes a Windows decision: It's true that networking and printing costs will be largely unaffected if existing practices are ratified as a corporate strategy, but not if other decisions are made. For example, an organization using mainly personal ink jet printers encounters page printing costs about five time higher than the same organization using the departmental lasers typically
found in Unix environments.

The centerpiece here is the Microsoft Great Plains software, comprising traditional financial management, manufacturing, and human resource management modules along with a series of custom products and third-party plug-ins to augment functions like contracts management, logistics, custom invoicing, and environmental reporting.

Look at this proposal in terms of where it leaves the company a year after the decision to proceed and both the risks and benefits are clear. The major risk, amounting in fact to a near certainty, is that costs will escalate far beyond the promised budget while the ERP implementation will drag out well
past the deadlines set in the plan.

The upside, however, is also clear. This conversion can reasonably be expected to produce a working system offering automated support for everything from quote generation to payroll and
bank reconciliations. Equally importantly, nobody gets fired for buying Microsoft.

The Unix architecture proposal

The Unix architecture proposal for Buldane is more radical, if not more costly. The company must:

  • Review and recommend long-term ERP options for the company, but base
    budgetary cost estimates on an Oracle suite covering financials, manufacturing, and inventory/logistics;
  • Replace all head office, plant, and branch PBX equipment while dramatically upgrading
    both internal and external networks to provide fully integrated voice,
    video, and data access to all sites;
  • Change the fundamental service architecture to a Unix/smart display model wherever permanent wiring
    exists or security is an issue and high-end desktop power is not required;
  • Give existing PCs to their users on condition those users first attend a Linux training class to be
    given by IT staff;
  • Replace about 140 engineering workstations, now a mixture of older gear, with new dual
    processor machines using Xeons, or Opterons if available, running Linux;
  • Replace existing Sun and Digital compute servers with two new Sun v880z visualization servers;
  • Replace all personal printers and fax machines with higher speed, PostScript based, departmental printers;
  • Replace all existing telecom and networking contracts with a single primary contract for national
    service plus a secondary contract with a different national network carrier providing head office
    links only;
  • Use the Sun Grid Engine to extend the power of the two V880s to include all of the wired engineering
    workstations and thus create a local supercomputer for modeling and other uses;
  • Make up project computing kits consisting of LinkSys DSL connectivity gear, Apple wireless
    networking, and Apple 17-inch PowerBooks for site engineering support.
  • Reduce net IT head count by about two-thirds over the transition period.
Component Configuration Quantity Unit cost1 Five-year total cost
Engineering workstations Dell 450n; Dual 2.8GHz Xeons, 1GB, 2 x 36GB, 15K, US3 disk, 20" LCD, Red Hat 9.0 140 $5,913 $1,241,730
AutoCAD upgrades From Autodesk V2 to LinuxCAD 140 $99 $20,790
Smart displays 17" Sun WANRay2 280 $787 $220,500
Project kits (base pack) 4 x Extreme base stations with antennae,
1 LinkSys 4-port DSL router, cables, 30-minute UPS
40 $1020 $61,200
Servers Sun V1280; 12 * 1.2GHz CPUs, 96GB, 2 x 36GB internal; dual 2GB/sec Fibre Channel controllers; 6120 2TB
RAID array;
4 $310,405 $1,862,430
Remote office systems Sun V240; 2 x 1GHz CPUs, 4GB, 2 x 36GB internal; 5 x 73GB External US3;
bundled with 20 Sun WANRay 17" displays ;
2 $29,900 $89,700
Avaya integrated telecom Two main switches; two subsidiaries, integrated voice/video/data; total of
790 handsets with cabling and installation (remote office handled by Telus;
1 $740,000 $1,110,000
Full ERP suite Budgetary estimate only, Oracle 90-day expedited, full suite; annual support is
1 $800,000 $1,600,000
IT staffing Staff; annual cost inclusive of benefits 12 $1,707,750 $8,538,750
Totals       $15,436,840
1Pricing generally from Avaya, Dell, and Sun Web sites as of July 1st, 2003

2Pricing for the Sun 17" WANRay and the upgraded V1280 is estimated. Product
and pricing announcements are expected this quarter.

The five-year total estimated cost for this proposal is about $15.5 million.

Notice that this proposal is more inclusive than the Windows one but still costs less. Nevertheless any
attempt to sell either proposal primarily on cost will fail, because the typical corporate decision maker treats both sets of numbers with the same offhand contempt.

Furthermore, any attempt to balance the comparison either by dropping about $1.6 million in communications-related costs out of the Unix proposal or adding the cost of comparable services to the Windows proposal will merely exacerbate the problem. Leave the costs out of the Unix proposal, and the money won't be in the budget when you need it. Try to put
them into the Windows proposal, and the typical decision maker will see, not the finger pointing at the
people fogging the Windows numbers, but the three pointing back at you.

The Unix alternative is much more difficult to sell in part because hardly anyone is familiar with it. To sell this proposal, you have to show it's significantly better, not just cheaper, than the
Windows alternative. There are many reasons to believe it is better, none of them
easily quantifiable. For example:

  1. Part of the core Unix message is choice. Once you buy the first Microsoft product, you don't have many choices for the rest. With Unix, there are always several choices, starting at the operating system level, with BSD, Solaris, and Linux currently leading the pack. Thus suggesting that a team be assembled to pick an ERP/SCM package that works for Buldane is a smart move because it emphasizes choice -- something the Microsoft people don't have.

    Setting the budgetary estimate based on adopting Oracle's financials and manufacturing suite on SPARC as the budgetary price model is also smart for two very different reasons. First, the
    upper limit at about $800,000 for a 90-day accelerated implementation project is both guaranteed and more expensive than alternatives from companies like PeopleSoft, thereby ensuring budget realism while not imposing constraints on the team making the choice. Secondly, and more importantly,
    looking at packages that would generally be thought of as massive overkill for the business ensures that
    customization, beyond getting the company name on reports, will mostly consist of deciding what
    not to use.

    Companies that invest heavily in customization based on adding or modifying functions or interfaces
    create barriers to their ability to adopt the vendor's subsequent releases.
    Most large scale SAP installations, for example, are now several releases behind vendor code because
    their customization resulted in new code and business processes
    which now prevent adoptation of new releases. Picking a package which has considerably
    more functionality than the company needs can reduce or eliminate the demand for customization and
    so avoids this trap.

  2. The Unix proposal includes significant networking upgrades and the replacement of existing PBX and
    related phone equipment with integrated integrated voice, video, and data stream systems from Avaya.

    Unix is fundamentally about communication over time and space. Giving people the ability to see and talk to each other without reference to
    distance or location improves communication and reduces barriers to successful interdepartmental
    collaboration. That it also has the potential to significantly reduce total telecom costs across
    the company is a bonus.

  3. Communication is what that oddball Mac recommendation is about too. Apple's Airport Extreme
    wireless base stations, bridges, external antennas and cards can provide 54Mb/sec
    communications across a four-building highrise construction site. Link one base station to a local server
    with an xDSL "phone home" connection to the nearest UUnet point of service and you have instant, high-speed communication and computation capabilities extending the corporate office to the construction

    The 17-inch LCD on the G4 laptop gives the on-site engineer the largest, brightest, on-screen X, PostScript, or PDF display available so when he
    wants to look at a drawing detail he can see it in daylight on the 29th floor -- and refresh the display
    from the latest engineering drawings at headquarters with little more than a mouse click.

    For these users communications is the issue, not computation or running office software.
    Buldane is fundamentally a structural engineering firm, the plant is just backwards integration. Thus Buldane's project engineers, who work on the building site during structural steel assembly, are there in part to make sure that the customer not only never regrets buying Buldane's products and services, but
    positively rejoices in the decision. This technology will help make that happen.

    As an example, imagine an engineer on the 19th floor of some structure
    where the elevator contractor has a placement problem. The quickest solution
    is to replace a support assembly on the central column. The contractor asks the engineer if it
    can be done. At that point the engineer needs access to drawings everyone can see,
    communication with an engineer at headquarters, and information from the plant on when replacement
    steel can be cast, rolled, and shipped. What he doesn't need is computation power. Without integrated communications this is a two-week design glitch holding up construction. With
    it, a Buldane engineer can give the customer an answer, a shipping schedule, and a temporary workaround, all in a couple of minutes.

Notice that none of these arguments are about "cheaper"; they're all about competitive position and customer service. That's also true of the centerpiece of the Unix proposal -- the suggestion that the smart display architecture be deliberately broken to create a local supercomputer in engineering with two high-end visualization engines and a grid accessing about 140 dual Xeons. This plan is irresistible to the people in metallurgical and structural design engineering. Obviously they get software continuity,
upgraded workstations, and the most powerful visualization engines available. Less obviously,
running Linux on that grid supercomputer means that it actually costs less to put in and run
than 140 single-processor Windows XP-based machines,
because AutoCAD is extremely expensive while LinuxCAD and continued use of the existing Fortran software are dirt cheap.

The critical issue, however, is competitive
advantage. The supercomputer shines here because it provides
an absolute killer sales tool that can be reasonably expected to
contribute positively to contract closings. It can allow Buldane's people, for example, to
show an architectural client an
Eiffel Tower, built to twice the height at half the weight using Buldane's beam products, reacting in real time
to a simulated magnitude 7 earthquake.

It is important too, to talk about how systems management will change between the two architectures. In the real world a company like this will have an extensive Wintel infrastructure. You make
your point by asking how they like it so far. On the Unix side, however, the fundamental issue is that
few people know what smart displays are, and fewer still have significant experience with them in a
well-managed environment.

Fixing the experience problem is relatively easy -- just set up a month-long demo using the kind of small
system package Buldane intends for its remote offices. This will give people some understanding
of the advantages these things offer.

Getting people to understand the management issue is far more difficult, and probably the fulcrum on which the final decision turns. Most people who use Unix in data centers
now centralize both processing and control. For the business architecture to work, however, only processing
should be centralized, with control distributed to the user community. Thus the real point of the demo isn't so much
to show that people can use the displays as it is to show that systems management can be trusted to listen to users
on daily operations.

Critical to that issue is user empowerment through knowledge. This is the last thing you want in a Windows
environment where a knowledgeable user is an uncontrollable threat to daily operations but developing and
encouraging such people has to be part of the
Unix management plan from the beginning. That's why I usually recommend that
everybody now using a PC
get an opportunity to attend a couple of seminars on loading and using Linux, and then take their
PC or laptop home as gifts from the company. Of course it's a bribe, but the training
builds trust relationships between users and the IT people giving the courses, and it gives users
the knowledge, and therefore the confidence, to exert more control over the systems they use.

This difference in management ideas needed to succeed with Unix reflects a fundamental philosophical difference between Unix and the Microsoft
products: Unix unifies organizations by making information and resources widely
available; Windows rebels against mainframe-style central control by fragmenting organizations.
The Unix view is of one computer, one network, pulling people together in one company;
the Windows mantra of "one man, one computer" creates islands of local control and automation,
potentially fighting each other for independence.

That's the real bottom line on the Unix vs. Windows cost issue for
business deployment; Unix is cheaper,
but competitive advantage in business isn't about
dollars, it's about harnessing the power of organization to realize
human growth and business opportunities.
Where Windows fragments organizations, Unix unites them, letting management focus
on opportunities, not cost containment.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 20-year veteran of the IT consulting industry.

- Write for us - and get paid! -

Click Here!