International Business Machines Corporation

Globalfoundries Inc.

Douglas R Miller

Michael A Blocksome

Joseph D Ratterman

Brian E Smith

Charles J Archer

Jeffrey John Parker

Sameer Kumar

Amith R Mamidala

Philip Heidelberger

Thomas M Gooding

Gabor J Dozsa

Michael K Gschwind

Valentina Salapura

Dong Chen

Paul William Coteus

Alan G Gara

Adam James Muff

Todd Edward Takken

Martin Ohmacht

Bruce Martin Fleischer

Alexandre E Eichenberger

Robert William Wisniewski

John A Gunnels

Mark Edwin Giampapa

Burkhard Steinmacher-Burow

Kathryn Mary O'Brien

Mark Gregory Megerian

Sameh W Asaad

Matthias Augustin Blumrich

Shawn Anthony Hall

Thomas Winters Fox

John Kevin O'Brien

Chen-Yong Cher

Barry Marshall Trager

Peng Wu

Charles David Wait

James Anthony Marcella

George L Chiu

Todd Alan Inglett

Krishnan Sugavanam

Gerard Vincent Kopcsay

Robert Matthew Senger

Noel A Eisley

Craig Brian Stunkel

Rudolf Adriaan Haring

David L Satterfield

Yutaka Sugawara

Kristan D Davis

Jose R Brunheroto

Michael B Mundy

Matthew Roger Ellavsky

Robert E Walkup

James Lee Van Oosten

Samuel J Miller

Brad Davis

Ruth Janine Poole

Ralph E Bellofatto

Peter Boyle

Brant L Knudson

Norman Christ

Kahn C Evans

William Maupin Stockdell

Alfred Thomas Watson, Iii

L Christian McDermott

A Multi-Petascale Highly Efficient Parallel Supercomputer of 100 petaflop-scale includes node architectures based upon System-On-a-Chip technology, where each processing node comprises a single Application Specific Integrated Circuit (ASIC). The ASIC nodes are interconnected by a five dimensional torus network that optimally maximize the throughput of packet communications between nodes and minimize latency. The network implements collective network and a global asynchronous network that provides global barrier and notification functions. Integrated in the node design include a list-based prefetcher. The memory system implements transaction memory, thread level speculation, and multiversioning cache that improves soft error rate at the same time and supports DMA functionality allowing for parallel processing message-passing.

Multi-petascale highly efficient parallel supercomputer

A system and method for enhancing barrier collective synchronization on a computer system comprises a computer system including a data storage device. The computer system includes a program stored in the data storage device and steps of the program being executed by a processor. The system includes providing a plurality of communicators for storing state information for a barrier algorithm. Each communicator designates a master core in a multi-processor environment of the computer system. The system allocates or designates one counter for each of a plurality of threads. The system configures a table with a number of entries equal to the maximum number of threads. The system sets a table entry with an ID associated with a communicator when a process thread initiates a collective. The system determines an allocated or designated counter by searching entries in the table.

Mechanism of supporting sub-communicator collectives with o(64) counters as opposed to one counter for each sub-communicator

A Multi-Petascale Highly Efficient Parallel Supercomputer of 100 petaOPS-scale computing, at decreased cost, power and footprint, and that allows for a maximum packaging density of processing nodes from an interconnect point of view. The Supercomputer exploits technological advances in VLSI that enables a computing model where many processors can be integrated into a single Application Specific Integrated Circuit (ASIC). Each ASIC computing node comprises a system-on-chip ASIC utilizing four or more processors integrated into one die, with each having full access to all system resources and enabling adaptive partitioning of the processors to functions such as compute or messaging I/O on an application by application basis, and preferably, enable adaptive partitioning of functions in accordance with various algorithmic phases within an application, or if I/O or other processors are underutilized, then can participate in computation or communication nodes are interconnected by a five dimensional torus network with DMA that optimally maximize the throughput of packet communications between nodes and minimize latency.

Methods, apparatus, and products for minimally buffered data transfers between nodes in a data communications network are disclosed that include: receiving, by a messaging module on an origin node, a storage identifier, a origin data type, and a target data type, the storage identifier specifying application storage containing data, the origin data type describing a data subset contained in the origin application storage, the target data type describing an arrangement of the data subset in application storage on a target node; creating, by the messaging module, origin metadata describing the origin data type; selecting, by the messaging module from the origin application storage in dependence upon the origin metadata and the storage identifier, the data subset; and transmitting, by the messaging module to the target node, the selected data subset for storing in the target application storage in dependence upon the target data type without temporarily buffering the data subset.

Minimally buffered data transfers between nodes in a data communications network

A parallel computer includes nodes, each having main memory and a messaging unit (MU). Each MU includes computer memory, which in turn includes, MU message buffers. Each MU message buffer is associated with an uninitialized process on the compute node. In the parallel computer, managing internode data communications for an uninitialized process includes: receiving, by an MU of a compute node, one or more data communications messages in an MU message buffer associated with an uninitialized process on the compute node; determining, by an application agent, that the MU message buffer associated with the uninitialized process is full prior to initialization of the uninitialized process; establishing, by the application agent, a temporary message buffer for the uninitialized process in main computer memory; and moving, by the application agent, data communications messages from the MU message buffer associated with the uninitialized process to the temporary message buffer in main computer memory.

Managing internode data communications for an uninitialized process in a parallel computer

Methods, systems, and products are disclosed for administering an epoch initiated for remote memory access that include: initiating, by an origin application messaging module on an origin compute node, one or more data transfers to a target compute node for the epoch; initiating, by the origin application messaging module after initiating the data transfers, a closing stage for the epoch, including rejecting any new data transfers after initiating the closing stage for the epoch; determining, by the origin application messaging module, whether the data transfers have completed; and closing, by the origin application messaging module, the epoch if the data transfers have completed.

Administering an epoch initiated for remote memory access

A shared address space on a compute node stores data received from a network and data to transmit to the network. The shared address space includes an application buffer that can be directly operated upon by a plurality of processes, for instance, running on different cores on the compute node. A shared counter is used for one or more of signaling arrival of the data across the plurality of processes running on the compute node, signaling completion of an operation performed by one or more of the plurality of processes, obtaining reservation slots by one or more of the plurality of processes, or combinations thereof.

Shared address collectives using counter mechanisms

Internode data communications in a parallel computer that includes compute nodes that each include main memory and a messaging unit, the messaging unit including computer memory and coupling compute nodes for data communications, in which, for each compute node at compute node boot time: a messaging unit allocates, in the messaging unit's computer memory, a predefined number of message buffers, each message buffer associated with a process to be initialized on the compute node; receives, prior to initialization of a particular process on the compute node, a data communications message intended for the particular process; and stores the data communications message in the message buffer associated with the particular process. Upon initialization of the particular process, the process establishes a messaging buffer in main memory of the compute node and copies the data communications message from the message buffer of the messaging unit into the message buffer of main memory.

Internode data communications in a parallel computer

Intranode data communications in a parallel computer that includes compute nodes configured to execute processes, where the data communications include: allocating, upon initialization of a first process of a compute node, a region of shared memory; establishing, by the first process, a predefined number of message buffers, each message buffer associated with a process to be initialized on the compute node; sending, to a second process on the same compute node, a data communications message without determining whether the second process has been initialized, including storing the data communications message in the message buffer of the second process; and upon initialization of the second process: retrieving, by the second process, a pointer to the second process's message buffer; and retrieving, by the second process from the second process's message buffer in dependence upon the pointer, the data communications message sent by the first process.

Intranode data communications in a parallel computer

A system and method for enhancing barrier collective synchronization on a computer system comprises a computer system including a data storage device. The computer system includes a program stored in the data storage device and steps of the program being executed by a processor. The system includes providing a plurality of communicators for storing state information for a bather algorithm. Each communicator designates a master core in a multi-processor environment of the computer system. The system allocates or designates one counter for each of a plurality of threads. The system configures a table with a number of entries equal to the maximum number of threads. The system sets a table entry with an ID associated with a communicator when a process thread initiates a collective. The system determines an allocated or designated counter by searching entries in the table.

Mechanism of supporting sub-communicator collectives with O(64) counters as opposed to one counter for each sub-communicator

Growing community of inventors

Albert Lea, MN, United States of America

Douglas R Miller