gWLM concepts / overview

Better use of the server capacity you already have

Typically, servers are set up with a single workload and ample reserve capacity to handle the peak demand of that workload. gWLM allows you to combine multiple workloads on a single server and make use of the reserve capacity.

Confidence that your mission-critical workloads get the resources they need

Even with multiple workloads on a server, you can ensure your mission-critical workload gets the resources it needs: gWLM automatically adjusts resource allocation, making it easy to share resources when they are plentiful, but also to dedicate those resources to workloads when there are spikes in resource demand.

Reduced system administration costs

With gWLM managing resource allocation, you can combine more workloads on fewer servers--reducing administration costs.

Workload

The collection of processes that are within a single compartment. The compartment can be an nPartition (npar), a virtual partition (vpar), a processor set (pset), or a Fair Share Scheduler (fss) group. gWLM manages a workload by adjusting the system resource allocations for its compartment.

Compartment

An npar, a vpar, a pset, or an fss group with its resource allocation being managed by gWLM.

Multiple compartments--of the same type--are grouped to form a shared resource domain, or SRD. The compartments all share the resources within the SRD. Each compartment holds a workload and can be in only one SRD. gWLM manages each workload by adjusting the resource allocation for its compartment.

Shared Resource Domain (SRD)

A collection of compartments--all of the same type--that share system resources. The compartments can be npars, vpars,psets, or fss groups. A server containing npars can be an SRD--as long as the npar requirements are met. Also, a server or an npar divided into vpars can be an SRD for its vpar compartments. Similarly, a server, an npar, or a vpar containing psets can be an SRD for its pset compartments. Lastly, a server, an npar, or a vpar containing fss groups can be an SRD for its fss group compartments.

gWLM creates SRDs when you use the Manage New Systems wizard, reached through the menu

Optimize Global Workload Manager (gWLM) Manage New Systems

Policy

A policy is a collection of settings that instruct gWLM how to manage a compartment's resources--and thus the resources available to the workload within the compartment. For example, a policy can indicate the amount of CPU resources a workload owns (and is guaranteed when needed) as well as how much of those resources the workload can lend to other workloads.

A single policy can be associated, or applied, to multiple workloads.

Mode

gWLM offers two modes: advisory and managed. Advisory mode allows you to see what CPU requests gWLM would make for a compartment--without actually affecting resource allocation. You can use this mode when creating and fine-tuning your policies. Once you are comfortable with your policies, use managed mode to have gWLM automatically adjust the resource allocations for your defined compartments.

You can only set the mode on the SRD level: All compartments within an SRD operate in the same mode, either advisory or managed.

Deploy

Enable gWLM control of an SRD.

Deploying an SRD in managed mode enables gWLM control of resource allocation within the SRD. For example, in an SRD based on a vpar that has psets for compartments, deploying an SRD in managed mode allows gWLM to actively migrate CPUs between psets.

When deploying an SRD in advisory mode, gWLM simply reports what the allocation would be--without actually affecting resource allocations on a system.

Undeploy

Disable gWLM's management of resources in a specified SRD.

If an SRD is in managed mode, undeploying stops the migration of system resources between compartments in the SRD. If the SRD is in advisory mode, gWLM no longer provides information on what requests would have been made.

HP-UX Hardware Partitions (npars)

A hardware partition, also known as an nPartition or npar, is a physical partition of a server, where each npar runs its own instance of the HP-UX operating system.

Using the HP product Instant Capacity, gWLM simulates the movement of CPUs between npars by turning off an active CPU in one npar then turning on a deactivated CPU in another npar in the same complex. Thus, the first npar has one less active CPU, while the second npar has one additional active CPU. (gWLM maintains the number of active CPUs, honoring the Instant Capacity licensing limits. As a result, no costs are incurred.) An npar must meet several requirements to be used as a compartment.

HP-UX Virtual Partitions (vpars)

A virtual partition is a software partition of a server or of a single nPartition, where each virtual partition runs its own instance of the HP-UX operating system. A virtual partition cannot span an nPartition boundary.

Processor sets (psets)

A processor set is a collection of processors (CPUs) grouped together for the exclusive access by applications assigned to that processor set. Each application runs only on processors in its assigned processor set.

Compartments based on psets are available on HP-UX and Linux. (On Linux, gWLM simulates psets using CPU affinity masks.)

HP-UX Fair Share Scheduler groups (fss groups)

A group of processes that has its CPU allocation managed by the Fair Share Scheduler that is available with HP-UX. A benefit of fss groups is their granularity: You can allocate fractions of CPU resources, rather than only whole CPUs, to the group of processes.