For non-specific (scratch) allocations, by default, MVS Device Allocation will first randomize across all eligible libraries and then once a library is selected, will randomize on the eligible devices within that library. In terms of the TS7700 Virtualization Engine, "library" is in reference to a composite library since MVS allocation has no knowledge of the underlying clusters (distributed libraries). z/OS also provides an alternate selection algorithm that will instead randomize scratch allocations across all eligible devices. This alternate algorithm is available at z/OS V1R11 and above, as well as z/OS V1R8 through V1R10 with APAR OA26414 installed.

The default algorithm "EQUAL" works well if the libraries under consideration have an equal number of online devices. For example, if two libraries are eligible for a scratch allocation and each library has 128 devices, over time, each library will receive approximately half of the scratch allocations. Likewise, if one of the libraries has 128 devices and the other library has 256 devices, over time, each of the libraries will still receive approximately half of the scratch allocations regardless of the number of devices in the library.

The alternate algorithm "BYDEVICES" will instead randomize scratch allocations across all devices. For example, if two libraries are eligible for a scratch allocation and each library has 128 devices, over time, each library will receive approximately half of the scratch allocations, similar to the "EQUAL" algorithm. However, if one of the libraries has 128 devices and the other library has 256 devices, over time, the library that has 128 devices will receive approximately 1/3 of the scratch allocations and the library that has 256 devices will receive approximately 2/3 of the scratch allocations.

With the default algorithm "EQUAL", there may also be times when device randomization within the selected library (or composite library) appears unbalanced either across subsystems or across clusters in a TS7700 Virtualization Engine (multi-cluster grid configuration running in balanced mode). As the number of eligible subsystems increases, the likelihood of this imbalance occurring also increases as the number of subsystems increases. If an imbalance is seen that impairs the overall throughput rate of the library, consider enabling the "BYDEVICES" algorithm.

For specific allocations (JES2-only environment), an ordered list of subsystems may be returned to MVS allocation. The environments that support an ordered list include the 3494 tape library and the TS7700 Virtualization Engine (Release 1.5 and higher). With the TS7700 Virtualization Engine, the list of subsystems returned is in cluster preference order. When an ordered list is returned, MVS allocation attempts to allocate from the first subsystem returned in the list. If an online device is not available within that subsystem, MVS allocation moves to the next subsystem in the list and tries again until a device is chosen. The default algorithm "EQUAL", supports an ordered list for the first seven subsystems returned in the list. After that, if an eligible device is not found, all of the remaining subsystems are considered equal. The alternate algorithm "BYDEVICES" removes the ordered subsystem limitation. With the TS7700 Virtualization Engine (Release 1.5 or higher), additional APAR OA30718 should also be installed prior to enabling the new "BYDEVICES" algorithm. Without this APAR, the ordered subsystem list may not be honored properly, causing specific allocations to appear randomized.