International Business Machines Corporation

Other

Matthew Richard Craig

Matthew J Kalos

Clint Alan Hardy

Lokesh M Gupta

Beth Ann Peterson

Scott Brady Compton

Kevin J Ash

Harry M Yudenfriend

Carol Santich Mellgren

Matthew G Borlick

Dale Francis Riedy, Jr

Richard Anthony Ripberger

Susan Kay Candelaria

John Glenn Thompson

Bret Wayne Holley

Roger Gregory Hathorn

Matthew John Ward

Joshua J Crawford

Micah Robison

John R Flanagan

Daniel Francis Casper

Gustav E Sittmann, Iii

Scott M Carlson

Louis William Ricci

Catherine C Huang

Edward Lin

Chung Man Fung

Mark P Bendyk

Qiang Xie

Tri M Hoang

Provided are a computer program product, system, and method for determining whether to process a host request by training a machine learning module. A machine learning module receives input information related to at least one of running tasks, mail queue messages related to host request processing, I/O request processing, and a host request received from a host system and produces output for the host request indicating whether to process the host request. A determination is made of output produced from the machine learning module based on the input information for a received host request. A determination is made of an expected output based on a measured system resource usage and an optimum system resource usage. The machine learning module is retrained with the input information, the output produced from the machine learning module, and the expected output and provided to determine whether to process a received host request.

Determining whether to process a host request by training a machine learning module

Provided are a computer program product, system, and method for managing swappable data structures in a plurality of memory devices based on access counts of the data structures. Data structures indicated as swappable are updated less frequently than most frequently updated data structures. Data structures not indicated as swappable are maintained in a first level memory device and not moved to a second level memory device. The first level memory device has lower latency than the second level memory device. Access counts are maintained for the data structures stored in the first level memory device that are indicated as swappable. Data structures are selected in the first level memory device having lowest access counts. The selected data structures are removed from the first level memory device and retained in the second level memory device.

Managing swappable data structures in a plurality of memory devices based on access counts of the data structures

Provided are computer program product, system, and method for managing data structures in a plurality of memory devices that are indicated to demote after initialization of the data structures. Indication is made to data structures to demote after initialization from a first level memory device to a second level memory device. The first level memory device has lower latency than the second level memory device. In response to completing initialization of the data structures in the first level memory device, the data structures indicated to demote after initialization are copied from the first level memory device to the second level memory device and removing the data structures indicate to move after initialization from the first level memory device.

Managing data structures in a plurality of memory devices that are indicated to demote after initialization of the data structures

Provided are a computer program product, system, and method for maintaining data structures in a virtual memory comprised of a plurality of heterogeneous memory devices. Access counts are maintained for a plurality of data structures stored in a first level memory device. A determination is made of data structures in the first level memory device having lowest access counts. The determined data structures are deleted from the first level memory device and retaining copies of the data structures in a second level memory device, wherein the first level memory device has lower latency than the second level memory device.

Maintaining data structures in a memory subsystem comprised of a plurality of memory devices

Provided are a computer program product, system, and method for determining feature settings for code to deploy to a system by training a machine learning module. A determination is made of an outcome of running system code on a system having configuration settings and feature settings of features in the system to enable or disable in response to the outcome. A machine learning module is trained to produce the feature settings indicating to enable or disable the features in response to input comprising the configuration settings of the system.

Determining feature settings for code to deploy to a system by training a machine learning module

Provided are techniques for selecting a disconnect by training a machine learning module. A machine learning module is provided that receives inputs and produces an output. The output produced from the machine learning module based on the inputs for the first I/O operation and an estimated amount of time to acquire resources for a first I/O operation is determined. An actual amount of time to acquire resources for the first I/O operation is determined. The machine learning module is retrained based on the inputs, the output, and the actual amount of time it took to acquire resources for the first I/O operation versus an estimated amount of time to acquire the resources for the first I/O operation. The retrained machine learning module is used to select one of disconnect from a channel, the logical disconnect from the channel, or the physical disconnect from the channel for a second I/O operation.

Selecting a disconnect from different types of channel disconnects by training a machine learning module

Provided are a computer program product, system, and method for prefetching cache resources for a write request from a host to tracks in storage cached in a cache. Cache resources held for a plurality of tracks in a write set are released before expected writes are received for the tracks in the write set. Cache resources for tracks in the write set are obtained, following the release of the cache resources, to use for expected write requests to the tracks in the write set.

Obtaining cache resources for expected writes to tracks in a write set after the cache resources were released for the tracks in the write set

Remote copy operations are performed to copy data from a primary storage controller to a secondary storage controller, wherein input/output (I/O) requests are received at the primary storage controller from a host both via a bus interface and a network interface while the remote copy operations are in progress, and wherein consistency groups are formed during the remote copy operations to copy the data consistently. Quiescing of I/O operations performed via the bus interface are performed while a current consistency group is being replaced by a next consistency group during the remote copy operations.

Management of data written via a bus interface to a storage controller during consistent copying of data

Provided are techniques for adjusting a dispatch ratio for dispatching tasks from multiple queues. The dispatch ratio is set for each queue of a plurality of queues. A number of Central Processing Unit (CPU) cycles used by tasks from each of the plurality of queues during the interval is tracked. A CPU high percentage is determined that indicates a percentage of CPU cycles used by high priority tasks. In response to determining that the CPU high percentage is below a high threshold, a new dispatch ratio is calculated that indicates an increased number of high priority tasks are to be dispatched, and the new dispatch ratio is based on the CPU high percentage, the high threshold, and a current dispatches high value. The increased number of high priority tasks are dispatched from the high priority queue based on the new dispatch ratio during a new interval.

Adjusting a dispatch ratio for multiple queues

A machine learning module receives inputs comprising attributes of a storage controller, wherein the attributes affect allocation of a plurality of resources to a plurality of interfaces. In response to a predetermined number of I/O operations occurring in the storage controller, a generation is made via forward propagation through a plurality of layers of the machine learning module, of an output value corresponding to a number of resources to allocate to an interface. A margin of error is calculated based on comparing the generated output value to an expected output value that is generated from an indication of a predetermined function based at least on a number of I/O operations that are waiting for a resource and a number of available resources. An adjustment is made of weights of links that interconnect nodes of the plurality of layers via back propagation, to reduce the margin of error.

Determine a load balancing mechanism for allocation of shared resources in a storage system by training a machine learning module based on number of I/O operations

A plurality of interfaces that share a plurality of resources in a storage controller are maintained. In response to an occurrence of a predetermined number of operations associated with an interface of the plurality of interfaces, an input is provided on a plurality of attributes of the storage controller to a machine learning module. In response to receiving the input, the machine learning module generates an output value corresponding to a number of resources of the plurality of resources to allocate to the interface in the storage controller.

Determine a load balancing mechanism for allocation of shared resources in a storage system using a machine learning module based on number of I/O operations

One general aspect of device reservation state synchronization in accordance with the present description, device reservation management logic ensures synchronization of reservation states of primary and secondary volumes of a mirror relationship in the event of a change in the state of the mirroring relationship such as achieving full data synchronization between the volumes. Other features and aspects may be realized, depending upon the particular application.

Device reservation state synchronization in data mirroring

Provided are techniques for cloud based store and restore with copy services. A store command to transfer data from one or more tracks of a volume to cloud storage is received. With track services, data for the one or more tracks in the volume is retrieved by emulating a host read. With a cloud data movement engine, the data for the one or more tracks is converted to data for one or more objects. With the cloud data movement engine, the one or more objects are stored in the cloud storage.

Cloud based store and restore with copy services

Provided are a computer program product, system, and method for managing cache segments between a global queue and a plurality of local queues by training a machine learning module. A machine learning module is provided input comprising cache segment management information related to management of segments in the local queues by the processing units and accesses of the global queue to transfer cache segments between the local queues and the global queue to output an optimum number parameter comprising an optimum number of segments to maintain in a local queue and a transfer number parameter comprising a number of cache segments to move between a local queue and the global queue. The machine learning module is retrained based on the cache segment management information to output an adjusted transfer number parameter and an adjusted optimum number parameter for the processing units.

Managing allocation and demotion of cache segments between a global queue and a plurality of local queues by training a machine learning module

Provided are a computer program product, system, and method for managing cache segments between a global queue and a plurality of local queues using a machine learning module. Cache segment management information related to management of segments in the local queues and accesses to the global queue to transfer cache segments between the local queues and the global queue, are provided to a machine learning module to output an optimum number parameter comprising an optimum number of segments to maintain in a local queue and a transfer number parameter comprising a number of cache segments to transfer between a local queue and the global queue. The optimum number parameter and the transfer number parameter are sent to a processing unit having a local queue to cause the processing unit to transfer the transfer number parameter of cache segments between the local queue to the global queue.

Growing community of inventors

Auburn Hills, MI, United States of America

Matthew Richard Craig