September, 1995

Newest IMAGE/SQL datasets tackle largest databases

Enhancement extends HP3000 reach to high-growth applications

HP3000 customers have been working to deploy very large databases (VLDBs) with IMAGE/SQL this summer, now that HP is shipping an upgrade that enables Jumbo datasets in the award-winning database.

VLDBs stand at the heart of an important growth segment for the HP3000, mail-order catalog management. Customer databases in excess of 30 million names are common for companies such as Time-Warner Cable, Hammacher-Schlemmer and other companies which have recently installed the HP3000. Such databases carry more than 20Gb of data, the commonly accepted starting point for a VLDB.

But until this year, IMAGE/SQL and its predecessors had a limit that restrained the practical size of HP3000 databases. No dataset could exceed 4Gb in an IMAGE/SQL, TurboIMAGE or IMAGE database. Applications which called for larger datasets couldn’t be easily managed on HP3000s.

The dataset barrier stood as a roadblock to HP’s march to make the HP3000 a mainframe alternative. The limitation was a function of the MPE/iX operating system, a bottleneck that’s been removed with the latest release.

By using the 5.0 Push release of MPE/iX, the HP3000 sews up to 10 chunks of 4-Gb datasets together, pushing the dataset limit to 40Gb. HP employed its new POSIX hierarchical file system naming feature in MPE/iX to extend the limit.

While HP admits that only 10 percent of its HP3000 customer base is pushing the old dataset limits, the new jumbo datasets have an impact on every site. HP’s IMAGE/SQL lab manager Jim Sartain said that expanding IMAGE/SQL into VLDB territory gives the database new growth opportunities in markets such as health care and mail order operations.

“There are some very successful software vendors that have been selling into major new HP3000 installations,” Sartain said. “They’re going after the top companies in respective industries, and those companies would have encountered the old dataset limits.”

Testing tools

Jumbo dataset tests have been underway for several months at some of the largest HP3000 installations in the world. At Current, Inc., a $300 million card and wrapping paper supplier and a division of check and form publisher Deluxe, a 60-Gb production database is planned for deployment later this year. HP system specialist Dave Jennings said his four-processor Series 995 system will support more than 900 online users. He’s preparing for the busiest period of Current’s sales year, the months which precede the year-end holidays, by testing jumbo datasets this spring.

Jennings reported that only a handful of tools have been revised to work with the datasets that are key to HP’s growth for the 3000. Current is using Suprtool from Robelle Consulting, and it was also deploying a version of Adager that handles jumbo datasets.

“If I didn’t have these programs, I’d be handcuffed,” Jennings said. “A few of our datasets are now 12 gigabytes. I couldn’t finish my job because one of my datasets filled up, and I didn’t have any tools to expand it. Adager is the only supplier [of an expansion tool] who’d commit to support jumbo datasets in a reasonable time frame.”

The short list of tools tested to work with the datasets is a result of the extensive changes in HP’s design. HP’s Sartain said the expansion of the datasets “is a major change to the [file system] structure, so it’s a major amount of work [for tools vendors]. And it’s a major amount of work with no obvious with no short-term financial benefit to them.”

While some HP3000 customers are buying database tools regardless of whether it supports the jumbo enhancement, Jennings said it wasn’t an optional feature for his operation at Current. HP didn’t hesitate to add the feature, either, according to Sartain.

“It really wasn’t an option, because we didn’t want to forego these big deals,” he said. “People who have been using it are very happy.”

Jumbo datasets, not prices

Neither of the HP3000 tool suppliers who support jumbo datasets today are charging extra for the feature, even though potential customers have budgets much larger than most HP3000 sites. Jennings said he’s procured about $500,000 of software for the two Series 995s and Series 937 HP systems since he arrived at Current. But the size of the prospective jumbo dataset user hasn’t triggered higher prices for tools yet.

“We don’t charge extra for the kind of native-mode RISC support Adager does, like dynamic detail dataset expansion and jumbo datasets,” said Adager’s Alfredo Rego. “We’re here to help our customers travel the Infobahn, not run them over.”

Other suppliers are working toward supporting the datasets. Omnidex, from Dynamic Information Systems Corp., is being tested both at Current and at Pilgrim Health Care, a health maintenance organization headquartered in Norwell, Mass. HP3000 director Jim Harding said he’s looking over a long list of programs essential to his 6-way Series 995 system.

“I’m still not overly confident that all of the third party products will work together with files larger than the 4-gigabyte limit,” Harding said. In addition to Adager, Suprtool and Omnidex, Pilgrim Health Care is using Vesoft’s MPEX, Quest’s Netbase, Cognos’ Quiz and Quick, Lund’s performance products, and Unison’s DCM Pak. Pilgrim is also using the Third Party Indexing (TPI) features for IMAGE/SQL from HP.

“ My greatest concern is with searching through datasets larger than 4 gigabytes with the indexed products,” Harding said. “Keeping databases shadowed with Netbase and indexed involving TPI has been difficult enough within the 4-gigabyte limit.”

The work at Pilgrim and Current involves some of the largest datasets in the HP3000 world. Harding said his medical claims and services application, AMISYS from Advanta, has one dataset with more than 20 million records, and another of more than 9 million records.

IMAGE/SQL and expansion plans

Customers using AMISYS and the MAC mail order software from Smith Gardner Associates are among the first to need jumbo datasets, but Sartain said HP plans to be releasing the enhancement to the full customer base this month. Sartain said no serious critical problems were reported across 30 beta sites testing the software this spring.

HP will make the enhancement available as part of its patch system, bypassing the delay of waiting for another full release of MPE/iX. But there are already discussions from the HP3000 community that a more thorough change will be needed before long - because 40-gigabyte datasets someday might not be large enough, either.

Wirt Atmar, reporting to the SIGIMAGE membership from this year’s Interex IPROF conference, said the jumbo datasets “are much more of a fix than a solution.

“If the pressure exists to maintain datasets that exceed 4 gigabytes, this 10-times extension will soon become unacceptable, too,” Atmar stated. “The relatively ‘small’ dataset size limit of 40 Gb is going to become a severe problem in the near-term future - simply because of the changing nature of what people are going to consider appropriate material to be stored in databases.”

As HP customers continue to fill IMAGE/SQL databases with business-critical data, Atmar and others say the best solution involves calls for reworking the MPE/iX file system. Atmar noted that MPE/iX still has room to grow, provided HP continues to be willing to invest in the work needed.

“Because the problem lies in the file system itself and not in IMAGE,” Atmar reported, “Craig Fairchild, the MPE file system architect, was asked during the [IPROF] conference, ‘Is it possible for the file system to be fixed?’ Craig said, ‘Anything’s possible.’ And that, too, was precisely the right answer.”