International Business Machines Corporation

Globalfoundries Inc.

Michael B Mundy

Thomas M Gooding

Michael E Aho

John Eric Attinella

Todd Alan Inglett

Andrew Thomas Tauferner

Samuel J Miller

Michael K Gschwind

Valentina Salapura

Dong Chen

Paul William Coteus

Michael A Blocksome

Alan G Gara

Joseph D Ratterman

Philip Heidelberger

Adam James Muff

Todd Edward Takken

Amanda Peters

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

Matthias Augustin Blumrich

Sameh W Asaad

Shawn Anthony Hall

Thomas Winters Fox

John Kevin O'Brien

Chen-Yong Cher

Barry Marshall Trager

Peng Wu

Charles David Wait

Sameer Kumar

James Anthony Marcella

George L Chiu

Krishnan Sugavanam

Gerard Vincent Kopcsay

Jeffrey John Parker

Patrick Joseph McCarthy

Robert Matthew Senger

Noel A Eisley

David L Satterfield

Douglas R Miller

Craig Brian Stunkel

Rudolf Adriaan Haring

Amith R Mamidala

Yutaka Sugawara

Kristan D Davis

Jose R Brunheroto

Matthew Roger Ellavsky

Gordon G Stewart

Robert E Walkup

James Lee Van Oosten

Ruth Janine Poole

Gabor J Dozsa

Ralph E Bellofatto

Peter Boyle

Brant L Knudson

Norman Christ

Kahn C Evans

Thomas A Budnik

William Maupin Stockdell

Alfred Thomas Watson, Iii

Brian E Smith

James C Sexton

Alan Jonathan King

Configuring compute nodes in a parallel computer using remote direct memory access ('RDMA'), the parallel computer comprising a plurality of compute nodes coupled for data communications via one or more data communications networks, including: initiating, by a source compute node of the parallel computer, an RDMA broadcast operation to broadcast binary configuration information to one or more target compute nodes in the parallel computer; preparing, by each target compute node, the target compute node for receipt of the binary configuration information from the source compute node; transmitting, by each target compute node, a ready message to the target compute node, the ready message indicating that the target compute node is ready to receive the binary configuration information from the source compute node; and performing, by the source compute node, an RDMA broadcast operation to write the binary configuration information into memory of each target compute node.

Configuring compute nodes in a parallel computer using remote direct memory access ('RDMA')

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

Collectively loading an application in a parallel computer, the parallel computer comprising a plurality of compute nodes, including: identifying, by a parallel computer control system, a subset of compute nodes in the parallel computer to execute a job; selecting, by the parallel computer control system, one of the subset of compute nodes in the parallel computer as a job leader compute node; retrieving, by the job leader compute node from computer memory, an application for executing the job; and broadcasting, by the job leader to the subset of compute nodes in the parallel computer, the application for executing the job.

Collectively loading an application in a parallel computer

Aggregating job exit statuses of a plurality of compute nodes executing a parallel application, including: identifying a subset of compute nodes in the parallel computer to execute the parallel application; selecting one compute node in the subset of compute nodes in the parallel computer as a job leader compute node; initiating execution of the parallel application on the subset of compute nodes; receiving an exit status from each compute node in the subset of compute nodes, where the exit status for each compute node includes information describing execution of some portion of the parallel application by the compute node; aggregating each exit status from each compute node in the subset of compute nodes; and sending an aggregated exit status for the subset of compute nodes in the parallel computer.

Aggregating job exit statuses of a plurality of compute nodes executing a parallel application

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.

Remote direct memory access ('RDMA') in a parallel computer, the parallel computer including a plurality of nodes, each node including a messaging unit, including: receiving an RDMA read operation request that includes a virtual address representing a memory region at which to receive data to be transferred from a second node to the first node; responsive to the RDMA read operation request: translating the virtual address to a physical address; creating a local RDMA object that includes a counter set to the size of the memory region; sending a message that includes an DMA write operation request, the physical address of the memory region on the first node, the physical address of the local RDMA object on the first node, and a remote virtual address on the second node; and receiving the data to be transferred from the second node.

Remote direct memory access ('RDMA') in a parallel computer

Messaging in a parallel computer using remote direct memory access ('RDMA'), including: receiving a send work request; responsive to the send work request: translating a local virtual address on the first node from which data is to be transferred to a physical address on the first node from which data is to be transferred from; creating a local RDMA object that includes a counter set to the size of a messaging acknowledgment field; sending, from a messaging unit in the first node to a messaging unit in a second node, a message that includes a RDMA read operation request, the physical address of the local RDMA object, and the physical address on the first node from which data is to be transferred from; and receiving, by the first node responsive to the second node's execution of the RDMA read operation request, acknowledgment data in the local RDMA object.

Messaging in a parallel computer using remote direct memory access ('RDMA')

In a massively parallel computer system embodiment, when receiving a message at a compute node from an input/output node, the compute node performs the steps of: obtaining a lock on a collective network device; checking a shared storage location for a message pending for a thread; if such a message is pending, receiving the message's remaining packets directly to a user's buffer, unlocking, and returning; if no such message is pending, receiving one packet from the network device; if the packet indicates that the message is for the thread, receiving the message's remaining packets directly to the user's buffer, unlocking, and returning; and if the packet indicates that the message is for another thread, updating the shared storage location with a thread id of the other thread, unlocking, waiting for a time out, locking, and repeating from the checking step. Accordingly, data copying is eliminated with an attendant performance benefit.

Providing improved message handling performance in computer systems utilizing shared network devices

Executing MIMD programs on a SIMD machine, the SIMD machine including a plurality of compute nodes, each compute node capable of executing only a single thread of execution, the compute nodes initially configured exclusively for SIMD operations, the SIMD machine further comprising a data communications network, the network comprising synchronous data communications links among the compute nodes, including establishing one or more SIMD partitions, booting one or more SIMD partitions in MIMD mode; establishing a MIMD partition; executing by launcher programs a plurality of MIMD programs on two or more of the compute nodes of the MIMD partition; and re-executing a launcher program by an operating system on a compute node in the MIMD partition upon termination of the MIMD program executed by the launcher program.

Re-executing launcher program upon termination of launched programs in MIMD mode booted SIMD partitions

Executing Multiple Instructions Multiple Data ('MIMD') programs on a Single Instruction Multiple Data ('SIMD') machine, the SIMD machine including a plurality of compute nodes, each compute node capable of executing only a single thread of execution, the compute nodes initially configured exclusively for SIMD operations, the SIMD machine further comprising a data communications network, the network comprising synchronous data communications links among the compute nodes, including establishing a SIMD partition comprising a plurality of the compute nodes; booting the SIMD partition in MIMD mode; executing by launcher programs a plurality of MIMD programs on compute nodes in the SIMD partition; and re-executing a launcher program by an operating system on a compute node in the SIMD partition upon termination of the MIMD program executed by the launcher program.

Executing Multiple Instructions Multiple Data ('MIMD') programs on a Single Instruction Multiple Data ('SIMD') machine

Growing community of inventors

Rochester, MN, United States of America

Michael B Mundy