Server savings

As he began building National Gypsum’s e-commerce site, Mike Brannon set up a server for order tracking, then one for PDF invoicing and another for patch testing. Launching separate servers for each additional function became habit forming, and costly, he says.

Three years and 53 servers later, Brannon has found himself with dozens of servers with underutilized processing power. And each box meant more equipment to manage, plus more downtime to deal with. “If you have to replace a disk drive, memory or a power supply, you’re down for that time,” he says. Moreover, many applications could have easily run on a 350-MHz Pentium III, but every time a hardware lease expired, he was forced to upgrade to a more powerful machine.

The senior manager of Internet technology has cured his bad habit with VMware’s ESX Server partitioning software, a server consolidation fix he projects will save his company $5,800 per month on hardware leases. Brannon retired the 53 HP DL360 servers, and replaced them with an eight-processor IBM X440 production server, and a four-processor HP server for testing.

VMware’s virtual machine software, which lets customers run multiple instances of an operating system and multiple applications on a single Intel processor, is one of several new tools that offer some form of server virtualization. Others include:

•  Software from Opsware (formerly Loudcloud) automates provisioning, change management, patch management, application deployment, and other server and application operations across multiple data centers.  Think Dynamics’ Think Control Suite creates a pool of server resources and uses policies to automate provisioning of those resources to meet unexpected demand shifts. The TopSpin 360 Switched Computing System provides a switch module that dynamically provisions virtualized memory and CPUs, storage and networking I/O, including 10-Gigabit InfiniBand connections.

• 

• 

The billion-dollar server automation market has grown in response to years of unbridled server proliferation, which drove customers to want to more efficiently manage their server resources, says John Humphreys, an analyst for IDC.

Different tools are a better fit than others, depending on the situation. For example, Think Control might be better-suited to a homogeneous Web farm, where the master code gets updated once and the new image of that operating system is replicated across hundreds of servers. On the other hand, Opsware automates and saves configuration changes at the application level and is well-suited to a complex mix of systems.

Users can deploy a variety of virtualization tools, says Galen Schreck, an analyst with Forrester Research: It’s not an either-or situation, but they need to calculate the risk involved. It’s conceivable with VMware that a server’s CPU could crash. “In some situations it might take out all the virtual computers that are running on the same CPU,” Schreck says.

Michael Mullany, senior director of product management for VMware, notes that vulnerability is mainly at the hardware level, for example, if the CPU overheats, or the server’s memory chip is bad, or if a network interface card or host bus adapter card fails. On the other hand, an operating system failure would not affect other virtual servers running on that same CPU. The physical memory that’s allocated to each virtual server is isolated below the operating system level, which stops events such as memory leaks, he says.

Humphreys concurs. “Pooling resources to allocate those to where the demand is at any given moment adds more risk. If the system fails, potentially more than one service is down.”

National Gypsum was well aware of the risks involved in deploying VMware, Brannon says. “Some guys get nervous about putting all of our eggs in one basket. Obviously, if that one basket crashes, you’ve crashed them all.” Eventually, Brannon expects to counter that risk by placing a load-balancing device out in front of two large ESX complexes.

Brannon now has 80% of his systems running on VMware. Most of his applications don’t require a lot of horsepower, so running VMware on an eight-CPU box saves hardware money. It’s easier than installing and managing half a rack of 1U boxes, and while he still needs to physically copy files to create software packages for provisioning or re-imaging servers, setup times have dwindled from days to hours.

While VMware virtualizes servers and CPUs, it doesn’t automate server configuration or patch management such as Opsware System 3 software. “Opsware can patch, upgrade and reconfigure the server in place without having to roll out a whole new image,” Schreck says. Companies with a lot of different hardware and different operating system revisions need a tool that will help them manage all that complexity, he adds.

Electronic Data Systems (EDS) is in the process of deploying Opsware in its global application-hosting centers, encompassing 50,000 server boxes running Windows, Linux and Sun Solaris. Larry Lozon, global service executive for the Web and application hosting business, is looking to increase response time to customer demand using Opsware. For example, EDS has used Opsware to provision 60 servers in three days, a process that previously took four weeks, Lozon says. EDS says it expects to save $100 million in operational costs over the next three years with Opsware.

Policy deployment

If you want to go beyond consolidating servers and automating system changes, Think Dynamics’ Think Control treats the server as a pool of virtualized resources that can be redeployed at any time, Schreck says. “It’s really more about workload management, where, based on demand for an application, you change the number of instances that are running. I can go from 100 to 300 Web servers if demand spikes.”

Tira Wireless deployed Think Control to handle the demands of its customers’ service-level agreements. “Predicting the load on a server is tricky. If there’s a spike in download traffic or performance degrades, we have to be able to switch, or add another box, to load balance or increase capacity immediately,” says Allen Lau, CTO for the wireless application publisher in Toronto.

Think Control is rules-based and detects when set thresholds are exceeded. It automatically configures all the necessary software, such as Web Logic or Apache TomCat, on a second server identical to the first box, Lau says. Provisioning a server and balancing the load takes just minutes, saving Lau the $100 per hour cost of hiring a consultant to manually install, configure and test servers. Tira also saves by not having to buy extra hardware to handle the 2% of the time that the spike is happening, he says.

Balancing the download needs of wireless carriers and portal needs of Java 2 Platform Enterprise Edition developers, Tira benefits from Think Control’s ability to toggle the use of its Sun Solaris boxes between load management, development and testing.

Brigham Young University is looking to improve the performance of its Web, application and database servers with the TopSpin 360 Computing System, a policy-based, multiterabit switching appliance that interconnects and virtualizes server resources to dynamically reprovision memory, CPUs, storage and bandwidth.

The Provo, Utah, school chose TopSpin because of its InfiniBand connection speed. Brigham Young plans to move to the Oracle 9I RAC database application, switching from large IBM S80s and HP 8410s to a clustered network of smaller 2U-high servers, and possibly blade server technology.

The bandwidth will be essential for managing the changing demands of the school’s data center, says Sorrel Jakins, chief engineer of operating systems. “The InfiniBand 10-gigabit-per-second bandwidth is important because if you get a bottleneck down in your database, it’s going to have a ripple effect through the data center.”

Brigham Young hopes to deploy TopSpin this year in conjunction with the move to the rack system. While the clustered environment will add failover capability and redundancy, the TopSpin virtualization will instantaneously save time and labor, Jakins says. It will let IT spread out performance loads, upgrade operating systems and reallocate resources without a lengthy reconfiguration time.

Jakins expects improved server utilization efficiency with TopSpin. “Managing and administering database systems with TopSpin will make it easier for us to build systems. Things are not as static as they once were,” he says.