| September 29, 1997 |
Distributed Systems: Back To The Middle
About 80% of IS managers say they plan to recentralize systems management or have done so
More than 60% of the 250 IS executives surveyed by
InformationWeek
over the summer say they plan to centralize distributed systems. Another 20% say they already have.
The reasons? In order of frequency, the survey respondents answered: easier management, lower costs, improved security, easier disaster-r
ecovery procedures, increased performance, and simpler storage strategies.
The move to centralize marks a decline in the client-server frenzy of the late '80s and early '90s. Back then, the attraction of cheap Mips (millions of instructions per second) caused nearly every user company to take servers from the glass house and move them close to business departments.
For some, that meant hundreds of servers
in every corner of the world. The Mips were indeed cheap, but by 1994, companies realized that their management costs had risen exponentially. Though users were very happy, client-server computing cost more money than mainframe computing. Also, IT managers found that data stored on distributed systems can be less secure than it is on mainframes. "Everyone got burned bad on client-server initiatives," says Jerry Peterson, a senior VP at Tandem Computers Inc. in Cupertino, Calif.
To combat this trend, IS executives are giving users what they want with efficient information access, but bringing the servers back to a central location. Among the benefits, according to an August report from Giga: improved systems management and recovery time, and lower WAN traffic due to the elimination of some server replication and WAN-based backup. The risk, says Giga's Rankine, is higher disaster-recovery costs as larger applications are exposed to a single disaster event. Nonetheless, return on investment due to consolid
ation is 15% to 35%, says Giga.
Among the companies moving back to centralized IT management are Chrysler, Kemper Insurance, Pepsi-Cola, and Proffitt's. Each has its own reasons. Chrysler is sticking with a strategy, kicked off in the late '80s, that calls for consolidation. Pepsi is taking ownership of distributors around the country. Proffitt's is sticking with a hardware platform but consolidating applications after the merger of five retailers. And while Kemper always kept Unix servers centrally stored, the company got tired of updating every server whenever there was a software upgrade or fix.
Chrysler, the Auburn Hills, Mich., automaker, last year began consolidating 160 Unix servers formerly scattered around Detroit. Chrysler has halved that number and says that as a result, it's saving $1 million a year in operational costs on the hardware alone. But so far, the company hasn't consolidated any of the 325 office-applications servers it has around the world. "Some companies tried to separate I
S personnel between mainframe and client-server computing, but we challenged our people to change their mental model and still deliver capabilities to the business," says Susan Unger, Chrysler's executive director of IS. "We're taking a hard look at the robustness and security in the client-server environment, and we're in the process of server consolidation."
At Pepsi, a division of PepsiCo Inc. in Purchase, N.Y., CIO Ken Harris says he saves seven figures annually after consolidating 23 Unix servers into one central server. "The consolidation lets us do the E-commerce once and the business process once," he says. Adds Dennis Johnson, Pepsi's VP of accounts receivable and E-commerce, "IS wants to make the implementation as seamless as possible, and the centralized environment makes that easy."
Nothing is easy for Proffitt's, a $2.3 billion retailer in Jackson, Miss. The company comprises five retailers that have merged since 1993. As a result, the company's systems include five versions of accounts
payable, three versions of general ledger, and six versions of merchandising. CIO Mark Goldstein is beginning to consolidate the applications, starting with accounts payable. "We're dealing with collections of disparate data with stovepipes based on tightly coupled CICS, VSAM, and DB2," he explains. Goldstein will stay with the mainframe, as he hopes to be in production on March 1 with an integrated accounts-payable system from Walker Interactive Systems in San Francisco.
At Kemper, in Long Grove, Ill., Dennis Dunlap, manager of midrange systems engineering, doesn't worry much about application integration. But he has watched the insurer's Unix servers multiply substantially, complicating operations. In Dunlap's six years at Kemper, the company has gone from six servers to 90, he says. Now Kemper is standardizing on IBM's RS/6000 SP multinode Unix system as a way to consolidate servers. Yet he sees the biggest cost savings in personnel. "We probably delay additions to staff with consolidation, because I'd
have to add a person to maintain every 20 servers in the old framework," says Dunlap. He says he should be able to add the equivalent of 30 servers to the SP before hiring anyone else.
While each of these companies has its own reasons for recentralizing, four common trends emerge. First, there are no logical walls in IS between mainframe and client-server computing. Second, the money and staff saved by consolidation are redirected to more strategic projects. Third, a business unit or function needs a specific application that runs best on a mainframe or large Unix server. Fourth, it's easier to maintain and manage a large single node, or at least manage multiple nodes, from a single workstation.
No Walls
So why the consolidation? "Businesspeople from the customer groups were drafted for part-time server maintenance that led to wasted time and inefficient work," says executive director Unger. "For customers, IS has got to be more like the water and electricity."
At Kemper, there may be too many Unix servers to suit managers, but at least they all sit right next to the mainframe in the data center. "We have 800 Mips in CMOS-based mainframes on the floor, but I'm just getting used to calling them servers," says Dunlap.
Similarly, at Proffitt's, CIO Goldstein maintains some Unix servers, but his problems haven't stemmed from platforms. "Our costs are multiple millions of dollars per year by not having integrated information," he says. "Our goal now is to achieve ROI in two years and develop continuous ROI so that I save more money
than the cost of putting in a new system."
Costs related to mainframe Mips may be decreasing because of cost-of-ownership realities. According to George Walsh, IBM's director of business hardware for the System/390, because Unix didn't knock out the mainframe-in fact, Unix became even more expensive to run and maintain-many IS executives will be slow to migrate away from any established mainframe operating systems-even to Windows NT.
Though consolidation isn't always a win for the mainframe, Walsh expects a lot from his division now. "We've been presented a golden opportunity in that CIOs have moved away from apologizing for the S/390 [mainframe]," he says. "390 is staying there while the business world is moving so fast, and it's hard to achieve competitive advantage and time-to-market while you're rehosting."
New Business
Proffitt's Goldstein wants to establish an Enterprise Integration Architecture, middleware for a new layer of abstraction. With the new combination, the retailer can go to new places. "We've never been able to compare our pricing and results to others in the industry, and we don't have a clue about gross margins today," he says. "We'll work out better deals with our vendors. They knew more about us than we did, but that's going to change."
Chrysler, too, will venture into uncharted territory. The combination of server consolidation, IS professional availability, and systems-management tools is opening up avenues. Beylin says Chrysler can communicate via the Web with all its suppliers-who provide 70% of Chrysler's materials-regardless of the system on the supplier's side. "We struggled for many years in provi
ding service and updating supplies," he adds.
The server consolidation has also let Chrysler achieve constant customer support. Its customers include dealers, and the engineering, finance, and sales and marketing departments. They all prefer the reliability brought by consolidation. "We've taken the time for a problem ticket down from four hours to less than 50 minutes," says Mary Daun, Chrysler's manager of strategic planning and application support. "We build in alerts with the applications themselves. The value in consolidation is understanding the entire process."
One feature in mainframes and large Unix servers that supports all these applications is the ability to deploy resources dynamically within the system. Downloading information from three different servers in three different regions means horrible response time, says Charles Foley, VP of server marketing for Amdahl Corp. in Sunnyvale, Calif. "The big system now becomes a logical server spread out over three partitions and lets me put the work where the processing is strongest," he says. "If today you tell someone `I'll get to the problem, I'll purchase more hardware, I'll get back to you in two weeks'-that's why the unemployment lines get temporarily long."
Sometimes a single application requires a big server. At both Pepsi and Proffitt's, the
financial application will come from Walker. "Our volumes make us unique in requirements," says Pepsi CIO Harris. "The Walker software works well, and they support it well."
Proffitt's knows about high volume: "We have hundreds of millions of transactions, each an item that gets bought and sold," says CIO Goldstein. "That's why we're going to a single version of AP with Walker in a Parallel Sysplex [IBM's mainframe cluster] configuration."
Walker's top executive would rather talk about the market trend taking place than tout his account wins. "This recentralization is not a technology movement, but a business requirement from multiple perspectives," says Leonard Liu, chairman and CEO of Walker. "Companies are looking hard to increase worldwide competitiveness, and they find out that instead of multiple operating units creating their own systems, it's more effective to have one shared service available to multiple units."
Liu says mainframes of a few years ago lacked the flexibility to run enou
gh applications and support enough data access. He says today's mainframes provide plenty of flexibility, while maintaining necessary security. "We should not confuse business need-to-know with knowing exactly what's happening at every level of a transaction,"
Liu cautions.
Safer Systems
David Gelardi, IBM's RS/6000 program director for DSS solutions and DBMS marketing, isn't ready to dismiss everyone who invested in the distributed individual servers a few years ago. "But," he says, "the prolif
eration of operating systems, databases, and applications made for difficult administration. Many IS executives are now saying, `At the time I made a good decision, but now it's a nightmare for me.'"
According to Gelardi, advances in technology just make things scarier. The SP gives customers the option to have 100 nodes, but they appear as a single system image, and each node can be fixed or updated individually while others remain online. When companies have 100 or more servers, business users are likely to want to update the applications in waves. This comes at a time when most companies can't afford a single hour of downtime. "It's like a lunar eclipse," says Gelardi. "All the planets must be lined up at one time."
At Pepsi, Johnson understands the systems-management issue well. "Twenty-three technically capable Unix servers means 23 databases and 23 EDI formats," he says.
But at least one industry observer says there's no mass migration from one platform to another. "SAP and Walker come in
and say, `You have a mainframe, and now you can put this state-of-the-art application right on it,' and it's a compelling argument for people who don't want to get involved with complicated distributed applications," says Joshua Greenbaum, a senior analyst with Hurwitz Consulting, an IT market research firm in Newton, Mass.
However, Greenbaum doesn't endorse a mass migration from client-server architectures back to the mainframe. "Consolidation makes sense for companies doing it on the same hardware or operating system," he says. Changing platforms leads to implementation headaches, Greenbaum warns.
Maybe so, but Amdahl's Foley says he understands why recentralization has become such a hot trend. He says the idea might be three years old, but says the pieces didn't come together until recently. "Between '92 and '96, people started dropping in the servers for new applications, and it started the whole mind-set for application servers," says Foley. "I love Unix, but these things are like bunnies. One
or two are cute, but there's no such thing as just a couple."
Whatever the reasons behind the timing, recentralization is a plus for IT managers. "The buzzwords are flexibility and change, not a rigid paradigm for applications," says Greenbaum at Hurwitz. "The customer is king now."
|