by Derek Slater

SYSTEMS ARCHITECTURE: IT-Rex? The Mainframe Survives

News
Nov 15, 200011 mins
Developer

Reader ROI

Learn why mainframes are still the engine of choice for many jobs

Steal successful techniques to staff your data center

Understand why Linux, of all software packages, can now be run on a mainframe

Curt Schumacher has a three-year plan to migrate off the mainframes at the Chicago Board of Exchange.

He’s had this three-year plan since 1993.

Schumacher is vice president of systems operations for the Chicago Board of Exchange (CBOE), an options trading exchange. As such, he is in charge of all of the infrastructure that keeps the Exchange’s systems running around the clock. The CBOE has made all sorts of upgrades and changes in its computing architecture since 1993, but still those three big Amdahl processors keep plugging away at the heart of the company’s core trading system. “I still state that in three

years we want to be off of these systems,” Schumacher says, “but it always comes back to the old disciplines of the data center and its reliability.”

Reliability, uptime, data integrity—those are the key strengths of that least cool of all computing platforms, the mainframe. And that’s the principal reason why many CIOs are standing in Schumacher’s shoes, keeping their big iron dinosaurs alive no matter how many Unix and Windows NT mammals are scurrying about underfoot. Even so, various wags and pundits have continued to shovel dirt on the mainframe’s grave for years. To pick one recent example, The Robert Frances Group, a Westport, Conn.-based research house, issued a “Duel for Data Center Dominance” report last May that trumpeted IBM’s loss of market share to big Unix vendors Sun and Hewlett-Packard.

IBM has renamed and rearchitected the mainframe and its operating system several times in an attempt to stem the tide—most recently announcing its next generation as z/OS running on the z900 platform. All well and good, but if the declining sales of IBM’s mainframes (and Amdahl’s and Hitachi’s withdrawal from the market) are any indicator, the tide hasn’t turned—yet.

After all, the Chicago Board of Exchange fits the profile of the classic mainframe holdouts: financial companies, airlines, telecoms. And even in these industries, mainframe users have yet to overcome two key obstacles—outrageous software licensing costs and staffing difficulties. But despite these challenges, other businesses might do well to reexamine the potential benefits of having a highly reliable, mature computing platform anchoring their e-commerce systems in the age of Internet availability demands. After all, if you’re in a new-economy gunfight, you should carry a big gun.

Size Does Matter

The bread and butter for the mainframe in recent years has been power processing—online transaction processing (OLTP) applications that have high transaction volumes and a rapacious appetite for system uptime and data integrity.

CheckFree does that kind of processing—and the core bill-paying application it runs on the mainframe is relatively new, not a musty Cobol dinosaur wheezing along on its last legs. CheckFree went public in 1996 and used its newfound cash to acquire two competitors. New CTO Ravi Ganesan, charged with consolidating the three bill-paying systems into one, had experience overseeing a mixed mainframe/Unix/Windows NT environment with 70,000 users in a previous position at Bell Atlantic. “Everybody knew my pager number, so I have a very good understanding of the manageability of that environment,” he says, laughing. Ganesan, who has since become the company’s vice chairman, and his crew at CheckFree designed a system that uses NT for a number of applications, but relies on roughly 1,700 MIPS (million instructions per second) of mainframe horsepower for the back-end processing.

It is noteworthy that the software CheckFree hosts on the S/390 is a consumer-facing application that is up and chugging 24 hours a day. “Our customers don’t care what runs the back end, whether it’s a mainframe or 5,000 Commodore Amigas. But people get very emotional when you’re moving their money, so dial-tone quality is very important.” says Ganesan.

Total Systems Services (TSS) also relies on big iron. TSS is a Columbus, Ga.-based credit card processor with two big data centers. The company operates five clusters (parallel sysplexes) of IBM mainframes and 50-plus terabytes of Hitachi Data Systems DASD storage. Last Christmas season, TSS processed over 300 point-of-sale credit card authorizations per second. Mainframe usage is not an ideological issue for TSS; the company intranet and other applications run on Windows NT servers. “But for the amount of processing we do against our cardholder accounts—our real core business systems—we don’t see moving away from the mainframe,” says Ben Hollek, senior director of technical services for Total Systems Services in Norcross, Ga.

In terms of sheer OLTP capacity, Unix servers have made great strides during the past several years. But the mainframe’s power curve has also risen. According to Peter McCaffrey, IBM’s program director for the S/390, the company’s first generation of CMOS-based mainframes could run up to six processors at 11 MIPS each. The current generation (G6 in IBM’s parlance) can run 12 processors at 201 MIPS each in a single box—plus the parallel sysplex architecture allows as many as 32 boxes to appear as a single image, bringing the total theoretical capacity to 12x201x32, or 77,184 MIPS.

Unfortunately it’s tough to compare those numbers against what Unix servers offer, according to IDC analyst Steve Josselyn (IDC is a sister company of CIO’s publisher, CXO Media). Certainly you can, in theory, staple together a zillion Unix boxes to handle the same number of transactions per second. The catch lies in the software that keeps those systems up and running. Or doesn’t. The mainframe environment has been banged on incessantly for several decades now, and it’s a high-availability platform. Getting the same level of reliability for Unix servers requires a lot of investment in man power and software. Analysts say the cost of maintaining Unix-based OLTP systems rises significantly when you go from 96 percent uptime to 99.9 percent uptime, and that is even more true of Windows-based systems (see “Deciding Factors,” CIO, Feb. 1, 2000). Years ago, only banks, airlines and telecoms needed five nines; today, the Net has created 24-hour demand for customer access to all kinds of information, from retail systems to inventory data.

It’s not that users don’t have their frustrations with advances in mainframe technology. For instance, IBM touts its ability to connect the S/390 directly to standard networking protocols like Ethernet, gigabit Ethernet and asynchronous transfer mode (ATM). But Ganesan says CheckFree had enough problems with that particular technology to dump it. Instead, CheckFree’s mainframe uses IBM’s ancient Systems Network Architecture (SNA) to connect to Unix-based RS/6000 servers, which then connect to the rest of the network via TCP/IP. “Our usage isn’t evenly distributed, and at the sort of [OLTP] scale we need, we have not been lucky with our TCP/IP stack on the mainframe. I think [IBM] needs to do some more work on that,” Ganesan says.

Similarly, IBM’s geographically dispersed parallel sysplex (GDPS) technology allows for mainframes 40 kilometers apart to operate as a single system; TSS’s Hollek says that distance limitation is still too small to provide real reliability benefits on the east coast. “It protects you maybe from a tornado, but not a hurricane,” which can wreak havoc over a larger area, he says. TSS, like nearly every company with sensitive data, mirrors its critical systems, but that’s for naught if both copies get wiped out by the same storm. As a result, TSS may experiment with a third data processing location to lessen its risk exposure to natural disasters.

Two Big Buts

Cognizant of these flaws, IBM appears committed to upgrading the platform, especially since its S/390 hardware and software still constitute a billion-dollar business for Big Blue. However, the mainframe faces two other significant real-world challenges that can’t be fixed by vendor R&D expenditure. One challenge is the old bugaboo of software licensing; the other, newer problem is an ever-more-acute lack of IS staff with mainframe management and programming skills.

Software licensing has been a thorn in the side of mainframe users for many years, with vendors charging high yearly usage fees based on the size of the processors the software is running on. Since many companies beef up their capacity every year, software costs rise with the tide. Users grudgingly admit that software vendors have made some concessions in recent years. TSS has opened negotiations with several software providers to explore alternative forms of contracts. “We continue to work on that, to at least allow our financial people to understand where our costs will be going,” Hollek says. “We’re not pleased with the costs, that’s for sure.” Schumacher says the CBOE spends about $7.5 million on software every year, of which more than half goes for mainframe software. “And if I get bigger engines, our nice pals will send me a bigger [software] bill for nothing,” Schumacher says, pointing out the most unpopular feature of mainframe software licensing policies.

Josselyn of IDC is still more blunt. “If you look at the third parties like Computer Associates and BMC, those guys have been very resistant to reducing software costs. I thought it was interesting when both of them recently pointed the finger at IBM for difficulties in their businesses, when they should have looked in the mirror.”

IBM’s z/OS and z900 announcement (in September 2000) included reformed licensing policies based on actual use rather than capacity. The announcement also touted support for this new pricing policy from CA, BMC and a raft of other key software vendors. How this policy shift will change the checks CIOs write remains to be seen. If the change is big, it could open the door for a dramatic upsurge in mainframe adoption.

Staffing difficulties are a more recent dilemma. Once upon a time, Cobol programmers, MVS (the predecessor to today’s OS/390 operating system) and SNA staffers were plentiful. Now, though, Windows NT/2000 and Linux have all the cachet, which means many new college graduates have neither experience with mainframe technologies nor the inclination to sit in the data center running tape backups for their first two years of employment. One of the TSS’s solutions to this problem is a joint program the company helped develop in cooperation with the state of Georgia. College students get their bachelor’s degree or associate computer science degree paid for by agreeing to work in a Georgian IS department for several years. TSS also helps provide instruction at the local school, Columbus State University. Hollek says the company has brought in about 10 technical employees through this program. However, that’s clearly not going to singlehandedly solve the staffing shortage.

Linux: Cool or Silly?

There’s more hope in this year’s announcement of Linux support on the S/390. At first blush, this might appear to be a desperate attempt to hitch the mainframe’s falling star to the meteoric rise of open source software. After all, if the maturity of OS/390 is one of the mainframe’s key selling points, why buy an expensive box and run a much less robust operating system? “I wrote a white paper earlier this year about Linux on the mainframe and some of the other folks here laughed, but it looks like there’s some real interest out there,” says Bill Claybrook, a research director at Aberdeen Group. That interest centers around the ability to run Linux in one partition of a mainframe and use it as a development platform for applications that could then be ported to run on an MVS partition. Linux is also popular for hosting Web applications, which again can benefit from the mainframe’s relatively bulletproof performance.

It may seem odd that the ability to run critical systems could depend on the “coolness” of their hardware platform, but that’s IS. And Linux, TCP/IP stacks and such technologies may help restore to the mainframe some measure of coolness. In fact, it’s possible the mainframe will even become hip again, with a sort of classy retro appeal like hamburgers, dry martinis and platform shoes. But in light of the Internet’s relentless demand for uptime, the mainframe’s same old selling points may be the wellspring of its resurgence for a long time to come.