In-Depth

Case Study: Hardware Consolidation Reduces Management and Maintenance Costs

With newer and more powerful SunFire systems—and fewer of them—The Chicago Tribute cut costs and implemented a more efficient network configuration.

• By Stephen Swoyer
• 03/23/2004

A prominent force in American journalism for 150 years, Tribune Publishing Co.—which publishes The Chicago Tribune, New York’s Newsday, and The Los Angeles Times, among other major dailies—was recently forced to confront some post-millennial growing pains.

The information systems that hosted some of the Chicago Tribune Co.’s most important applications—including advertising, circulation, and database marketing programs – were more than half-a-decade old. What’s more, the maintenance costs associated with supporting them were growing more costly with each passing year. As if that wasn’t enough, Tribune’s aging hardware assets were under-utilized and under-available, which prevented the company from easily and inexpensively adding applications.

The Rx for the newspaper’s application and availability woes was dispensed by a quartet of companies: AT&T, EMC Corp., Nortel Networks, and Sun Microsystems Inc. Chicago Tribune tapped a pair of 72-processor SunFire Enterprise 15000 servers from Sun, optical switches from Nortel, new storage from EMC, and about four miles of black fiber from AT&T to build a scalable and available datacenter infrastructure.

When it goes live this summer, the daily’s new datacenter will reduce its ongoing management and maintenance costs, and, more importantly, will enable a new active-active wide-area clustering configuration that enhances both application availability and disaster recovery.

Growing Pains

Chicago Tribune’s aging data center infrastructure was anchored by a pair of Sun Ultra Enterprise 10000 (E10K) servers, the 64-way behemoths that Sun first introduced more than seven years ago, along with a dozen or so other systems, including 14-way Sun UltraEnterprise 5500 boxes. The daily also tapped a mainframe system to support its marketing database and other workloads.

One impetus for the upgrade, says Darko Dejanovic, vice president and chief technology officer with both Tribune Publishing and Chicago Tribune, was the cost of maintenance for the E10Ks and other assorted Sun boxes, which were becoming prohibitively expensive to support. “Consolidation also allowed us to cut our maintenance costs. As you know, some of these E10Ks, et cetera, the vendor—any vendor, really—they start charging you major amounts of money for coverage as they get older,” he confirms.

At the same time, he explains, the newspaper’s UltraEnterprise 5500s were under-powered, under-utilized, and—because of their distributed nature—expensive to manage. “We actually used to have much more free capacity on our individual systems, and the reason for that is they peak at different points. Our editorial content system peaks at night, the circulation system peaks at different points, as well as advertising, so [the new infrastructure] allows us to leverage more capacity so we don’t have so many peaks,” Dejanovic says.

Chicago Tribune’s former infrastructure also supported an active-passive clustering configuration, which helped to augment application availability, but which was also comparatively expensive. “Because we’ve got active-active [with the new configuration], we’re eliminating some licensing issues associated with active-passive, where you’ve got to pay for a separate [software] licenses even though it’s not being used” because it’s deployed as a passive node in a cluster, Dejanovic explains.

Ultimately, the shortcoming of the newspaper’s active-passive clustering configuration proved to be the tipping point for its data center upgrade. In most active-passive clustering implementations, after all, failover from one application to another can take several seconds or minutes, which makes for a far-from-seamless recovery.

“We asked ourselves, ‘Could we put a number of these applications on two 15Ks, through a cluster, and have an active-active?’ Which not too many people are running,” Dejanovic says. “This would accomplish a few things. First, it would give us full active-active, a highly available disaster-recovery situation. It would also reduce costs by eliminating [clustering] licensing issues, cut down our maintenance agreements, and give us a refresh of our technology across all of our boxes, and reduce the number of boxes [that we have to manage].”

Here Comes the Sun

In late 2002, Sun announced that its Sun Cluster 3.0 service could support active-active clusters with up to 200 KM of distance between nodes. An active-active topology helps to ensure that users at one site typically experience only a very minimal disruption as they fail over to another site, says Marty Robins, senior director of industry and product marketing with Sun. “It allows them to mirror their operations in a way that gives them survivability of their data center with minimal impact to users [in the event of an application failover],” he asserts.

At the time, Sun formed a partnership with Nortel to market the latter’s Optera 5200 DWDM switches as part of its “Enterprise Continuity” service offering. That relationship continues to this day, occasionally involving another partner—AT&T—which in this case provided the black fibre backbone between Chicago Tribune’s clustered sites. Although Sun typically markets its own StorEdge solutions as part of its Enterprise Continuity service, Tribune tapped new CX600 storage from EMC, which it clustered by means of Veritas Software Corp.’s Volume Manager.

Last December, The Chicago Tribune received its new SunFire 15Ks, each populated with 72 processors and 140 GB of memory. Although it doesn’t plan to move the bulk of its application infrastructure over to the new systems until later this summer, Dejanovic says that Tribune has already designed its next-generation data center infrastructure. It plans to place the two SunFire systems at locations approximately three to four miles apart from one another, as well as partition its applications into seven Solaris domains, which Dejanovic says will help to ratchet up utilization and increase overall efficiency. The newspaper’s systems host a combined 38 TB of data, 14 TB of which is currently stored in one location, 24 TB in another.

All told, the newspaper will consolidate from 12 down to two Sun boxes, which Dejanovic says should account for most of the upgrade-related ROI. “The major ROI for us came from the fact that when we looked at individualized boxes that we would have to buy to refresh the system, it was going to cost much more, and when we looked at the maintenance contract we’d have to carry in order to do that, it was costing us much more,” he confirms.

The extra horsepower of the new SunFire systems will enable The Chicago Tribune to introduce new workloads as well. “We are going to run the content, the editorial systems, parts of the advertising system … and from the perspective of the tests that we ran, and the announcements we’ve done, we’re eliminating a lot of the overhead that we used to have,” Dejanovic comments. “Ultimately, as we look at and get more comfortable with this technology, we’ll be looking to expand [into new applications], certainly we want to take advantage of that [additional scalability].,” In addition, he notes, the other dailies in the Tribune Publishing Group are watching Chicago Tribune’s wide-area active-active clustering implementation with interest. “We can actually split those boxes further [than three or four miles], so maybe the distance between Newsday or the Hartford Courant is a couple of hundred miles or less, and this could allow us to have quite some distance between business units and run it this way."

Whither the mainframe? Never fear, says Dejanovic, who says that Big Iron will remain in the newspaper’s infrastructure, although The Chicago Tribune plans to move its marketing database over to the new SunFire systems. Even in this respect, Dejanovic says, there’s an upside, as it lets Tribune reallocate its mainframe capacity for other workloads: “That allows us to free up the mainframe here in Chicago and consolidate on to others in LA and elsewhere.”