Managing Servers in the Cloud
Managing server loads is a complex endeavor in any environment, and even more so when virtualization and cloud computing comes into play. There are, of course, myriad solutions for managing server loads. Virtually Speaking: Load balancing is tricky stuff, but it gets even more complex when the workloads are a mix of real and virtual.
On Monday, Librato (which up until recently went by the name Evergrid) unveiled version 2.0 of Librato Load Manager, a product designed to deliver heterogeneous (read: physical, virtual or both) application workload management in x86 environments, without modifying applications or operating systems.
Ameet Dhillon, vice president of marketing at Librato, described Librato's mission as seeking to enable diverse workloads to run together by balancing resource consumption to achieve business goals, especially during peak resource usage periods
The holistic approach begins with an examination of how the application behaves. This helps ensure the policies put in place are the correct ones.
If holistic is one underpinning, heterogeneity is the other. Dhillon said that customers don't want a solution that will work for only part of the infrastructure. They want to be able to deploy across the entire business, and thus virtual and physical and multiple operating systems.
In that vein, this new version supports Windows, in addition to Linux, regardless of whether the OS is a guest or host.
Because of its "wrapped at the user level" it's also hypervisor-agnostic, Dhillion said. This capability to monitor, manage and control without having to modify is its biggest value he explained.
Like the typical load manager, Librato Load Manager monitors and controls computing resources: CPU, memory, network I/O and storage I/O.
Because it is policy-based, Load Manager is designed to bring in user data, which enables it to offer dynamic rebalancing so enterprise applications are allocated the appropriate system resources to achieve optimal performance and predictable service levels.
Unused capacity is dynamically reassigned, which means servers don't sit idle and resources are maximized.
Dhillon cited "sponge" consolidation as one use case. In a sponge consolidation, a server can have a dedicated app, and a second app that "soaks up" remaining utilization resources. The admin would allocate 100 percent of the resources to the primary app. The "Sponge" app is guaranteed for zero, unless primary is not using them. Sponge apps, is typically "compute and batch" apps that can run in off hours, he explained.
Although such usage easily lends itself to a virtualized environment, they wouldn't necessarily have to operate in virtual machine environments.
In general, "workload management, is complementary to virtualization," Dhillion said, so it's not a vast leap in logic to assume that heavily virtualized environments will consider a solution of this ilk to keep things humming.
Of course, many large shops likely have a Tivoli or Openview deployment already in place. These, Dhillon explained, "don't offer the level of granularity our customers need." They are at a much higher level, "other than being in the app itself, we're as close as you can get," Dhillon noted.
Librato Load Manager 2.0 is expected to begin shipping December 1 of this year.
In addition to adding support for Windows Server 2003 and 2008 environments, the new release adds monitoring capabilities and the provides users with a comprehensive view of application resource requirements for more accurate capacity planning.
Load Manager can also be used for capacity planning purposes. Admins can use it to determine how much resources an application will consume if left unchecked to enable them to better allocate resources.
Amy Newman is the managing editor of ServerWatch. She has been covering virtualization since 2001.