International Business Machines Corporation

Intel Corporation

Sap Se

Adnan M Agbaria

Kalman Zvi Meth

Su-Hsuan Huang

William Joseph Ferrante

Patrick Francis Caffrey

William Gole Tuel, Jr

Anton Prenneis

Demetrios K Michailaros

Jose Eduardo Moreira

Vijay Krishnarao Naik

Ehud Cohen

Nir Naaman

Somdeb Majumdar

Gidon Gershinsky

Konstantin Shagin

Edi Shmueli

Lior Storfer

Marcel Nassar

Xin Wang

Tristan Webb

Ron Banner

Evren Tumer

Logic may quantize a primary neural network. Logic may generate, by a secondary neural network logic circuitry for a primary neural network logic circuitry, quantization parameters. The primary neural network logic circuitry may comprise a primary neural network with multiple layers trainable with an objective function. Each of the multiple layers of the primary neural network may comprise multiple tensors. The secondary neural network logic circuitry may comprise one or more secondary neural networks trainable with the objective function to output the quantization parameters to the tensors.

Methods and arrangements to quantize a neural network with machine learning

Techniques and apparatus for providing graph compression structures for graph information are described. In one embodiment, for example, an apparatus may include at least one memory, at least one processing circuitry, and logic, coupled to the at least one processing circuitry, to access graph information comprising a plurality of nodes, define a unique index for each of the plurality of nodes, determine whether each of the plurality of nodes has at least one neighbor node, and generate a graph compression structure comprising an entry for each of the plurality of nodes having at least one neighbor node and an adjacency list comprising an array of neighbor nodes of each entry.

Techniques for information graph compression

According to some embodiments, it may be determined that a first field in a first table is to be linked to a second field in a second table. A computer processor may then automatically create a new entry in a linked field table, the new entry including a linked field identifier, a source value, a source link referring to the second field in the second table, and a link counter containing a value indicating that one field refers to the second field in the second table. The first field in the first table may then be updated to refer to the new entry in the linked field table.

Linked field table for databases

A method for migrating a virtual machine (VM) in a computing environment is provided. The method comprises receiving a request to migrate a VM executing on a source host to a destination host; defining a recovery point to which the VM is restored during recovery from a fault; and iteratively copying a memory of the source host associated with the VM to the destination host. During the copying, the original state of each page in the memory is preserved. At some point, the VM suspends executing on the source host, copies state information associated with the VM to the destination host, and resumes executing on the destination host. If a fault is detected on the source host, the VM is restored to the recovery point using preserved information.

Live migration of virtual machines in a computing environment

Deallocating resources allocated to a network communications session, including extrapolating an expected location of a selected node that belongs to a communications path of a network communications session, determining if the extrapolated location is beyond a predefined range from a neighboring node that belongs to the communications path of the network communications session and that immediately neighbors the selected node along the communications path, and deallocating any resources allocated to the network communications session if the extrapolated location is beyond the range.

Efficient deallocation of network resources based on network node location extrapolation

A checkpoint of a parallel program is taken in order to provide a consistent state of the program in the event the program is to be restarted. Each process of the parallel program is responsible for taking its own checkpoint, however, the timing of when the checkpoint is to be taken by each process is the responsibility of a coordinating process. During the checkpointing, various data is written to a checkpoint file. This data includes, for instance, in-transit message data, a data section, file offsets, signal state, executable information, stack contents and register contents. The checkpoint file can be stored either in local or global storage. When it is stored in global storage, migration of the program is facilitated. When a parallel program is to be restarted, each process of the program initiates its own restart. The restart logic restores the process to the state at which the checkpoint was taken.

System of performing checkpoint/restart of a parallel program

Method of performing checkpoint/restart of a parallel program

Program products for performing checkpoint/restart of a parallel program

A complete and consistent set of checkpoint files is captured and identified for use in restarting a parallel program. When each process of a parallel program takes a checkpoint, it creates a checkpoint file. The checkpoint file is named, and a part of that name includes a version number for the checkpoint file. When a parallel program is to be restarted, each process of the parallel program identifies its most current valid checkpoint file. It provides the version number of this file to a coordinating process. The coordinating process then decides which version of the checkpoint files is valid and complete for all of the processes of the parallel program participating in restart. Once this version number is determined, it is forwarded to the processes and the processes restore themselves using its corresponding checkpoint file having that particular version number.

Capturing and identifying a complete and consistent set of checkpoint files

A checkpoint of a process is taken in order to provide a consistent state of the process in the event the process is to be restarted. When the process is restarted, the process is placed in the state it was in when the checkpoint was taken. However, there are times when certain information has changed since the last checkpoint and it should not be restored. For example, if a process is restarted on a computing unit different from the one in which the checkpoint was taken, then environment variables associated with the new computing unit, such as the name of the computing unit, are not to be restored to the values they were before the checkpoint. In order to prevent this, one or more environment variables are adjusted, after the process is restored to the reflect the state they were in prior to restoration.

Restoring checkpointed processes including adjusting environment variables of the processes

A checkpoint of a process is taken in order to provide a consistent state of the process in the event the process is to be restarted. When the process is restarted, the process is placed in the state it was in when the checkpoint was taken. However, there are times when certain information has changed since the last checkpoint and it should not be restored. For example, if a process is restarted on a computing unit different from the one in which the checkpoint is taken, then various attributes associated with external data referenced by the process (e.g., an address to an external function/variable) might be different than at the time of the checkpoint. Similarly, even on the same computing unit, if the functions and/or variables are reloaded, the attributes may be different. Since these new attribute values are needed for the restarted process, they should not be restored. In order to prevent restoration of the attributes, only part of the Data Section is restored. The portion of the Data Section including the attributes remains intact.

Restoring checkpointed processes without restoring attributes of external data referenced by the processes

Growing community of inventors

Musmus, Israel

Adnan M Agbaria