Management:
The Missing Piece of the Puzzle

Finding tools to virtualize and consolidate your data center is easy. What's harder is finding the tools to properly manage this new virtualized world.

By: Robert L. Scheier

Building the next-generation data center is one thing. Managing it will be another. Much of the technology to virtualize and consolidate servers, networks and storage is here today or on the way. What will take longer, according to vendors, customers and analysts, is automating that technology to reconfigure itself so the most critical applications or database queries run the fastest. Only when that automation is possible will data centers deliver on the promise of lower-cost IT capacity that businesses can tap, like a utility, as it's needed.

IT managers face "a big spider web of understanding your transactions, and where they're running, and how they're running," says an IT manager at a national retail chain who requested anonymity. For example, "I just did a big purchase of an IBM enterprise service bus. Now, I've got to see inside this service bus, and see where the holdup is. Is it application code at the Web services level? Is it MQSeries [messaging middleware]? If you don't continually try to automate, [this work] will bury you."

Bill Randall, director of MIS infrastructure at restaurant chain Red Robin International Inc., says he hasn't yet seen an automated tool "that's going to make a better decision than some of my engineers," especially when it comes to correlating information coming from different parts of the IT infrastructure. "I'm not seeing any tools out there that are able to sift through that amount of data" and automatically determine the root cause and proper fix for a problem, Randall says. In fact, most of what research firm Gartner Inc. calls "real-time infrastructure" products are "specific to a particular problem, such as horizontally scaling applications for a particular platform, and have limited or no visibility of the end-to-end business transactions," according to a March 2006 report. The key standards and supporting products to enable true end-to-end dynamic management are at least several years away, say customers and industry observers.

The obstacles facing every customer will be unique, and so will the timetables companies follow for successfully managing the next-generation data center. Here are some of the most common challenges and how some key vendors are trying to meet them.

Beyond Cost Cutting

What's been driving the move to the next-generation data center? So far, it's mostly been cost savings. Companies still recovering from the dot-com bust are looking for ways to postpone or avoid the purchase of new servers, storage or software and minimize the size of their IT staffs. For example, Randall says EMC Corp.'s VMware virtualization software has cut the time needed to deploy a new server at Red Robin's Greenwood Village, Colo., headquarters from three to four hours to 20 minutes. Configuresoft Inc.'s Enterprise Configuration Manager (ECM) is also saving him four to six person-days per quarter by automating the quarterly reports he must provide to department heads about which users have permissions for accessing or changing data in various systems.

At Polaris Industries Inc., a manufacturer of snowmobiles, all-terrain vehicles and motorcycles in Medina, Minn., capital budgets have been de-creasing every year. "We can't add staff, but there's more pressure to add services," says Dean Buker, manager of network and systems operations. By virtualizing Hewlett-Packard Co. server blades with VMware, since early 2005 Buker has slashed the number of physical servers at Polaris from 120 to 70 and reduced the annual purchases of new servers by about 70%.

The corporate investment bank at Wachovia Corp. last year began using DataSynapse Inc.'s FabricServer for its transaction-oriented applications. Tony Bishop, senior vice president and director of product management, says FabricServer has delivered 50% better performance at an 85% cost savings. It has also allowed Wachovia to reallocate processing power among multiple data centers in multiple locations as the trading day progresses around the world.

As impressive as those numbers are, companies are demanding more than just savings. They want data centers to provide the flexibility and adaptability that will improve customer service and speed the introduction of products and services. In a May 2006 Computerworld/CIO survey, nearly as many IT executives cited new capabilities as the main reason to change their data centers (60%) as those who cited cost cutting (63%).

Data center managers are looking toward virtualization, consolidation and automation. Virtualization means making multiple virtual devices appear as a single logical device, or making a single physical device function like several physical devices. Consolidation means reducing the types or amount of hardware or software in the data center to cut real estate, power, licensing or management costs. Automation means reducing the need for human intervention to manage, troubleshoot or optimize the data center.

In general, consolidation and virtualization are easiest to achieve, while automation — specifically, automated management — is toughest.

Virtualization Buffet

There are several flavors of server-level virtualization, and that market is dominated by VMware. The obvious cost benefits have resulted in the explosive growth of VMware, Microsoft's decision to include a hypervisor in its next-generation Longhorn server and the development of the open-source Xen hypervisor for Linux.

IDC analyst Stephen Elliot and other observers say that as virtualization becomes ubiquitous, what will distinguish virtualization software is its ability to provide security, management, provisioning and the recording of CPU usage. These records will allow IT to charge business units for the resources they use — and to measure the value IT delivers to the business.

Among the companies offering such services are Marathon Technologies Inc., whose software links multiple Windows servers in a virtualized, high-availability environment; Leostream Corp., whose software manages server virtualization products from VMware and Microsoft; and PlateSpin Ltd., whose PowerConvert product automatically distributes servers and data among Windows virtual servers. In June, an upgrade to VMware's server virtualization software added new power management capabilities, a distributed file system that allows users to pool heterogeneous storage arrays into virtual volumes and a distributed resource scheduler that automatically redistributes virtual machines among the physical servers as application demands change.

PolyServe Inc. in Beaverton, Ore., performs virtualization at the file-server level with separate file-serving and database "utilities." With PolyServe, says CEO Mike Stankey, "customers can literally drag and drop the database onto a bigger or smaller server" and build policies dictating when performance slowdowns will trigger such a move. PolyServe complements, rather than replaces, proc-essor-level virtualization such as that provided by VMware, he says. That's because VMware and similar virtualization products don't automatically reconfigure those applications so backup and other functions work in the virtualized environment.

Work in Progress

Storage virtualization promises savings by increasing the utilization of existing disk space and easier management of fewer disk pools, as well as business benefits through improved information sharing. However, a lack of standards and immature technology have limited its deployment. Many storage virtualization products work best or work only within arrays from a single vendor. Virtualized servers sometimes don't work well with virtualized storage because currently, every virtual machine in a server must share the same external storage resources and a single set of addresses in a Fibre Channel storage-area network. This makes it harder for administrators to create separate zones for development, test and production virtual machines; prioritize SAN resources for mission-critical applications; and quickly move a virtual server to another physical machine in case of a hardware failure.

A new standard called N-Port ID Virtualization, or NPIV, allows a single host bus adapter (HBA) to use multiple addresses within a storage network. Now available only on IBM zSeries mainframes, HBA vendor Emulex Corp. is working to expand NPIV to other platforms.

Some virtualization is possible at the network level through products such as Cisco Systems Inc.'s Catalyst Switching Series. These products allow the partitioning of a single physical network into many virtual networks, which can be configured for the performance or security needs of specific applications or networks.

Where's the Holdup?

One of the big challenges is looking deep within complex, multitiered or service-oriented applications to see what hardware or software is causing a slowdown. CoreFirst Transaction Workload Management software from OpTier Ltd. in New York is one of the strongest players in this field, according to several customers, who declined to be identified.

Unlike other products, Op-Tier's CoreFirst examines queries within databases to find those that use too many CPU cycles, too much memory or too much of the server's I/O subsystem, says the IT manager at a leading retailer. It can also prioritize workloads so that a query from a cashier looking up the price for a product, for example, gets higher priority than a query about last month's sales.

A leading health care provider in the Midwest deployed CoreFirst to troubleshoot performance problems within a complex, fat-client customer-service application that gathers data from multiple other applications. "We can take our application and almost break it down by the business function" an employee is using, such as the time required to look up a customer record, says the company's director of technology support, who declined to be identified.

Virtualization may simplify some functions, but it complicates others, says Dennis Moreau, chief technology officer at Configuresoft. Configuresoft's ECM, for example, can help ensure that new virtual servers meet corporate and regulatory guidelines for security and access control, are properly configured, and have the proper versions of software and the appropriate security patches. By the middle of next year, Configuresoft plans to give users the ability to check the status and configuration of instances of an operating system that exist on a server but are not yet in operation.

The more tiers developers build into their applications, and the more data sources they touch, the more flexibility managers will want in changing — on the fly and automatically — the performance of those components so they better match business needs.

David Miller, a director in the technology and infrastructure practice at systems integrator Avanade U.K. Ltd., used PolyServe to reduce the number of "passive" (or hot spare) servers for a large database deployment from 500 to 125. PolyServe also helped reduce fail-over time from five minutes to 30 seconds. The next step, he says, is for products such as PolyServe to perform such fail-over automatically as applications require it, rather than requiring an administrator to initiate the change.

Systems administrators are reluctant to give up a lot of control to automated tools. Randy Wood, a technical support manager in BellSouth's Communications Systems Group in Atlanta, uses the Network Automation System from Opsware Inc., but only to view alerts from customer networks. "It's unlikely we would let the [software] just execute [a] change" without first contacting the customer, he says. A customer might have violated network routing policies, says Wood, but only to troubleshoot a problem — something the software would have no way of knowing.

The Value-Adds

Some companies turn to IT automation software not to manage their systems, but to monitor their compliance with regulations governing things such as access to critical data. That's the case at restaurant chain Red Robin, which uses Configuresoft's ECM to ensure the proper configuration of 45 Windows servers at its headquarters. The company also uses ECM to instantly change passwords in the service accounts in its Windows Active Directory domains, a process that used to take 48 hours, says Randall.

But it will be several years at least before a single "dashboard" exists, observers say. They advise IT managers to standardize the products they buy and the way they configure them to give automated tools a chance to work.

Scheier is a freelance writer in Boylston, Mass. Contact him at rscheier@charter.net