Load Balancing Using the Leaky Bucket Mechanism
Load Balancing Using Spill and Fill
Load Balancing between Magnetic Libraries
The software system includes many features that help you to balance the operations between available storage resources in the CommCell. For example:
This document describes how to configure the MediaAgent software to distribute the load between multiple resources.
Using the leaky bucket mechanism each resource will be utilized until the threshold established for the resource is reached. When the thresholds are reached, the system automatically uses the next available resource.
Consider the following examples:
This example illustrates how load balancing can be accomplished between 2 MediaAgents that share a library.
In this example, 2 MediaAgents (MediaAgent 1 and MediaAgent 2) are configured to share a library with 4 drives. Such a configuration would result in the following data paths:
To perform load-balancing between the 2 MediaAgents, you can allocate 2 drives for each of the MediaAgents. This can be done by as follows:
This ensures that jobs automatically switch-over to the next available resource, when the thresholds are reached.
In the above example, (assuming that several jobs are running concurrently at a given time) the first two jobs will use the resources in data path 1, while the third and fourth jobs use the resources in data path 2. Subsequent jobs in the queue will be routed to a data path with an available resource, as soon as a resource is freed.
This example illustrates how load balancing can be accomplished between a tape library and a magnetic library.
In this example, MediaAgent 1 has a magnetic library configured, while MediaAgent 2 has a tape library. Such a configuration would result in the following data paths:
To perform load-balancing between these 2 MediaAgents, in such a way that at any given time 2 jobs write to the magnetic library, while also using all the drives in the library. This can be done by as follows:
In the above example, (assuming that several jobs are running concurrently at a given time) the first two jobs will use the resources in data path 1, while the next four jobs use the resources in data path 2. Subsequent jobs in the queue will be routed to a data path with an available resource, as soon as a resource is freed.
Note that at any given time, the maximum number of jobs will be within the established threshold. (i.e., 2 for the magnetic library and 4, which is the maximum or the total number of drives in the tape library.)
The spill and fill method equally distributes the jobs among available resources.
The following example illustrates the spill and fill mechanism between tape and magnetic libraries:
In this example, 2 MediaAgents have been configured with tape/optical libraries, while 2 more MediaAgents have 2 magnetic libraries configured. Such a configuration would result in the following data paths:
To perform load-balancing between these 4 MediaAgents, in such a way that at any given time 2 jobs write to the magnetic library, while using all the drives in the tape/optical libraries. This can be done by as follows:
This ensures that the jobs are distributed equally among all the MediaAgents.
See Also: Load balance (Spill and fill) between Data Paths.
You can perform a simple load-balancing between 2 mount paths in a magnetic library, without using alternate data paths. This can be done using the Mount Path Allocation Policy in magnetic libraries and mount paths.
The following example illustrates this mechanism.
In this example, MediaAgent A has a magnetic library containing 2 mount paths. To perform load-balancing between these 2 mount paths, in such a way that at any given time 5 jobs write to the magnetic library, you can:
In the above example, (assuming that several jobs are running concurrently at a given time) the first five jobs will use mount path 1, and the next five jobs will use mount path 2. Subsequent jobs in the queue will be routed to either one of the mount paths, as soon as a mount path is freed.
Once the maximum allocated writers are used, the system will constantly check for an available resource, and as soon as one is freed the next job in the queue will be automatically routed to use that resource.