International Business Machines Corporation

Saravanan Devendra

David Alan Hepkin

Rajalakshmi SrinivasaRaghavan

Venkatesh Sainath

A method for metering accelerator usage in a computing system includes receiving, from an accelerator, a set of metering rules, the metering rules including a unit of accelerator usage and a tariff per unit of accelerator usage. The unit of accelerator usage corresponds with an amount of work performed by the accelerator in executing data processing operations. The method may further include receiving on a first bus from a partition of the computing system a command block, the command block having data processing operations executable by the accelerator. The method may additionally include calculating, based on the command block and the set of metering rules, an accelerator usage value for the partition. The method may then include providing, on a second bus, the accelerator usage value.

Metering accelerator usage in a computing system

A system, and computer usable program product for modeling memory compression are provided in the illustrative embodiments. A subset of candidate pages is received. The subset of candidate pages is a subset of a set of candidate pages used in executing a workload in a data processing system. A candidate page is compressible uncompressed data in a memory associated with the data processing system. The subset of candidate pages is compressed in a scratch space. A compressibility of the workload is computed based on the compression of the subset of candidate pages. Page reference information of the subset of candidate pages is received. A memory reference rate of the workload is determined A recommendation is presented about a memory compression model for the workload in the data processing system.

Modeling memory compression

A method for modeling memory compression is provided in the illustrative embodiments. A subset of candidate pages is received. The subset of candidate pages is a subset of a set of candidate pages used in executing a workload in a data processing system. A candidate page is compressible uncompressed data in a memory associated with the data processing system. The subset of candidate pages is compressed in a scratch space. A compressibility of the workload is computed based on the compression of the subset of candidate pages. Page reference information of the subset of candidate pages is received. A memory reference rate of the workload is determined. A recommendation is presented about a memory compression model for the workload in the data processing system.

Growing community of inventors

Karnataka, India

Saravanan Devendra