International Business Machines Corporation

King Shing K Lui

Leho Nigul

Dana Lea Price

Paul R Bastide

Kuan Feng

James Irwin Knutson

Marlin R Deckert

Dana M Duffield

Junfeng Liu

Zhichao Su

Embodiments for accelerator allocation and reclamation for deep learning jobs in a computing cluster. Metrics are recorded of each accelerator of a set of accelerators allocated to a deep learning job including computing a gain of computational power by an additional allocation of new accelerators and computing a cost of transferring data among the new accelerators and the set of allocated accelerators. Ones of the new accelerators are allocated to the deep learning job or ones of the set of allocated accelerators assigned to perform the deep learning job are reclaimed upon determining an optimal accelerator topology by comparing the gain of computation power and the cost of transferring data.

Smart accelerator allocation and reclamation for deep learning jobs in a computing cluster

Computer-implemented methods and systems are provided for recommending one or more applications to users connected to one or more social networks. The computer-implemented method may include determining frequency of interaction between the users connected to the one or more social networks, analyzing behavioral patterns of the users connected to the one or more social networks, the behavioral patterns derived by considering at least the frequency of interaction between the users, and enabling at least one service provider to recommend one or more applications to one or more of the users based on the monitored behavioral patterns.

Software recommendation services for targeted user groups

A method, system, and computer program product for managing a managing a middleware architecture is disclosed. As such, a plurality of specification levels can be managed concurrently within a same runtime. An application manager and a set of runtime execution stacks are established to form a middleware. The application manager is configured to establish an interface to the set of runtime execution stacks. The runtime execution stacks are configured to run independent of one another using representative specification technology runtimes. A request to start an application is received. The application includes an application manifest. A particular runtime execution stack of the set of runtime execution stacks on which to run the application is determined. The determination is made based on the application manifest and the representative specification technology runtimes. Using the particular application runtime execution stack, the application is started.

Managing middleware using an application manager

A fix delivery system receives situation data representative of a problem, analyzes the received situation data to determine whether the problem associated with the situation data matches a known problem, and responsive to a determination that the problem associated with the situation data does not match a known problem, conducts problem determination to identify the problem. The fix delivery system further receives a developed cumulative fix for the identified problem, builds an installable cumulative fix for the identified problem using the situation data, and responsive to a determination that the installable cumulative fix meets predetermined requirements and the installable cumulative fix is a new installable cumulative fix, saves the new installable cumulative fix in a storage system and sends the new installable cumulative fix to a requester.

Growing community of inventors

Unionville, Canada

King Shing K Lui