Case Study: Managing the virtual realm
Be prepared to say goodbye to many of your tried-and-true techniques as you step into virtual-server management.
By Denise Dubie, Network World | InfoWorld | Published: 09:00, 29 August 2006
Managing virtual servers requires IT managers to think outside the box - literally. System administrators trained to check the health and availability of physical servers now must figure out how to distinguish between what's happening at the physical level with what's going on in the virtual realm. Obviously, they must understand the many facets of virtual servers. Then they must couple that with knowledge of how to fine-tune each virtual machine as it resides with others on one hardware box. Simply put, running a virtual server environment calls for a new management mind-set, says John Hinkle, CIO at Trans World Entertainment, a US-based operator of retail music stores, such as Strawberries, Coconuts and FYE.
"Managing virtual servers is not highly complex, but it is quite different from traditional server management. There aren't one or two tricks that work in any situation; you have to learn how to look at a multi-system environment. It's easy to think one or two physical servers are one or two servers, but now in reality they can be so much more," he says.
Hinkle speaks from the experience of having collapsed three data centres into one and, in the process, virtualising four OS/400 partitions, one AIX partition and 10 Windows servers on an IBM iSeries 570 server. His team uses IBM management tools to monitor the overall infrastructure, and software from SolarWinds to monitor the physical and virtual servers' uptime and performance. At times the transition from physical to virtual servers was "a bit bumpy" as the team encountered new management requirements, he says.
Related Articles on Techworld
For example, the team learned that although management tools can deliver the same types of reports for physical and virtual environments, different actions are needed in response to an alert. In a virtual environment, for instance, systems administrators must dig deeper into an alert - perhaps using more advanced tools - to determine whether it relates to one virtual machine or affects the entire physical box. This ambiguity about the source of a problem could lead to alerts not being taken seriously enough, or could result in lengthy repairs causing server downtime, Hinkle says.
"On a traditional, stand-alone box, the server has a disk array. In a virtual environment, the disk is all shared, so the problem may not be on your virtual machine," he says. "You need to determine what is going on across the environment and the shared resources and pinpoint where the problem is," he says.
In a virtual server environment, a system administrator has to think of all the applications and processes that depend on the virtualised resources residing on the physical box, he adds. "In a virtualised environment, there are things, such as reboot and apply patches, you just can't do in the middle of a production day, because you are altering the underlying system that all the virtual machines touch," Hinkle says. "You have to think differently, and [that has been] our biggest challenge. It would be easy to work like you always did and not take advantage of the new infrastructure."
Indeed, says Charles King, principal analyst at research firm Pund-IT, "virtualisation requires IT operations staff to have a certain level of sophistication. And it requires technical savvy for managing the systems, especially in a large, cross-platform enterprise environment."
Reinventing systems management
Pinpointing performance problems inside the partitions in a virtual environment requires tools that can pick up on faults and errors beyond the hardware to the virtual machines inside, others agree.
At Bowdoin College, for example, deploying VMware ESX Server virtual servers helped the systems team eliminate about one-third of the college's physical servers and cut potential costs for new systems by more than $350,000. Keeping the virtual environment running without incident, however, required an investment in updated management technology, says Mitch Davis, CIO at the school.
"Despite how easy it may seem, deploying, maintaining and managing virtual servers require knowledge and familiarity with the underlying systems," Davis says.
The IT team realised it wouldn't get the necessary in-depth details about its virtual environment if it stuck only with its existing management tools: Nagios, an open source application that pings systems for availability, and HP Insight Manager, software agents that monitor the physical health of servers. By augmenting these tools with VMware's VirtualCenter management software, an ESX Server add-on, the IT team says it partially closed the monitoring gap.
"We were managing our virtual servers individually, and those interfaces started to add up. We bought VirtualCenter, and now we can access all the servers from a single interface," says Tim Antonowicz, a Bowdoin system administrator.
"Virtual management is really reinventing systems management in some ways, because it encompasses such a broad set of functions," says Tony Iams, vice president and senior analyst at research firm Ideas International. "The technology should include the ability to create, migrate and delete virtual machines; to monitor for performance; and should incorporate automation."
At Cars.com, BladeLogic provided the management capabilities Edward Christensen, director of technical operations, needed for his virtualised server environment. The online automotive company uses VMware to virtualise servers on HP boxes in its development and quality-assurance environments. It needs to have a "fast refresh cycle," Christensen says, and that requires virtualisation. It also requires a fast way to make sure systems are configured and operating up to preset policies. For that, Christensen tapped BladeLogic, which he already had been using for server management.
"BladeLogic does a compare and points out when a configuration is different, so we can better understand why one virtual server takes up more resources or keeps crashing quickly," he says. "We need an alert when changes are made so we can know what they are, instantly."
Virtual management in the works
Fortunately, with the New Data Centre in mind, several vendors are attempting to tackle the unique challenge of managing a virtual world. These include system-management specialists, including BladeLogic, Altiris and Opsware; virtualisation providers, such as VMware, which in June updated VirtualCentre with more advanced features; and start-up specialists, such as Cassatt. The latter company, a Network World start-up to watch for 2004, reports on all the operating systems, applications, servers and virtual machines within an environment. This spring, Cassatt updated its software to help manage virtual machines from VMware, Microsoft and the Xen open source project.
Among bigger vendors, heavyweights such as BMC Software, CA, HP, IBM and Microsoft - the latter three of which provide virtualisation products - are not to be counted out. They're adapting their traditional systems management tools to handle virtual machines.
For example, BMC is applying its Business Service Management (BSM) strategy - which relates business processes and services to the underlying IT infrastructure's performance - to a virtual environment. BMC Performance Assurance Suite for Virtual Servers provides capacity management for VMware ESX Server. It understands how business applications are performing within VMware ESX virtual machines and identifies hot spots - abnormal activity that can be corrected by administrators before it affects the business service, BMC says.
CA offers Unicentre Advanced Systems Management software, which automates the allocation of server capacity on demand with dynamic brokering, the company says.
HP late last year updated its Integrity Essentials product line with a feature that assesses the expected growth in virtual server workloads and suggests how to maximise server utilisation levels. In addition, HP Global Workload Manager, a component of HP Virtual Server Environment, helps enterprise IT shops monitor workloads based on policies. The tool, updated last fall, migrates resources automatically from idle servers to busy ones to give applications the CPUs they need.
Also last fall, IBM upgraded its Tivoli Workload Scheduler and Tivoli Intelligent Orchestrator tools so they now allocate batch workloads across available virtual resources based on prescheduled jobs. That means systems managers can run jobs alongside transactional application workloads whenever there is available capacity, eliminating the need to run off-hour batch jobs.
Finally, Microsoft this spring released an updated management pack that helps users manage Virtual Server installations. The software, which works with Microsoft Operations Manager, lets users monitor servers running Virtual Server and the virtual machines running as guests on those servers. The company also introduced System Centre Virtual Machine Manager, expected to go into beta by the end of this summer, for managing virtual images and provisioning workloads.
Virtually the beginning
Despite these efforts, virtual server management tools are still immature, industry watchers say. The technology isn't fully baked, says Adam Lord, a system architect at Bowdoin. "I'd like to see more real-time monitoring features. And I'd like to be able to drill down from the higher view of the virtual environment using that interface into the actual server," he says.
For now, enterprise IT shops should consider management products that include automated server provisioning and configuration, says George Hamilton, director of Enterprise Computing and Networking at Yankee Group. And they should require vendors to provide integration between physical and virtual server management tools, he says. They need to employ management tools specific to virtual machines on Day 1 of their virtual server deployment, he adds.
"Existing management tools will not tell the entire story - they may tell you about the physical resources, but they may not give accurate virtual-machine performance information," Hamilton says. "And manual provisioning and configuration can't scale as the number of virtual machines increases."