Cool and Steady Wins the Exascale Race

By Osman Sarood, Laxmikant Kale & Esteban Meneses

November 12, 2013

Power, energy and reliability present major challenges to HPC researchers in their endeavor to build larger machines. As we approach the exascale era, both hardware and software designers need to account for these challenges while optimizing performance. The Parallel Programming Laboratory (PPL) at the University of Illinois at Urbana-Champaign (UIUC) has been actively working on meeting these challenges by leveraging the adaptive runtime system of the Charm++ programming model.

Current petascale machines have Mean Time Between Failures (MTBF) that can be anywhere from a few hours to days. Some reports predict exascale machines will have an MTBF in the range of 35-40 minutes. Intriguingly, past research describes a relation between a processor’s temperature and its reliability: failure rates double with every 10C increase in temperature. Our work applies this relationship between processor temperature and reliability by restraining processor temperature, thereby reducing the frequency of faults and consequently improving application performance in fault-prone environments.

Next week at SC13 we will present and analyze the costs and benefits of improving reliability through temperature control driven by Dynamic Voltage and Frequency Scaling (DVFS). Improved reliability helps not only by directly decreasing failures; it also allows the code to checkpoint less frequently, decreasing overhead. However, it comes at a cost of slower processors and increased load imbalance.

By restraining processor temperatures, we can empower the runtime system to set the expected failure rate of the system, adjusting it within a feasible range. Our control strategy lets each processor work at its maximum frequency as long as its temperature is below a threshold parameter. If a processor’s temperature crosses the maximum threshold, it is controlled by decreasing the voltage and frequency using DVFS. When the voltage and frequency are reduced, its power consumption will drop and hence the processor’s temperature will fall.

When DVFS adjusts frequencies differently across the cores in a cluster, the workloads on those cores change relative to one another. This can significantly degrade performance of a tightly coupled parallel application, where processors synchronize after a time step before proceeding to the next time step. We mitigate the resultant timing penalty with a load balancing strategy that is conscious of the difference in speeds for different processors.

Our load balancing strategy, based on overdecomposition and object migration, uses the Charm++ adaptive runtime system to increase processor utilization. It analyzes the current load of each processor according to its new frequency and determines if it is overloaded or under-loaded. Once this decision is made, our scheme intelligently exchanges objects from overloaded (hot) processors to under-loaded (cold) processors to balance load. Temperature checking and corresponding load balancing can be invoked at user defined intervals.

One twist in this work is that different applications vary both in how hot they will make processors at a given frequency, and in how their performance is affected by different frequencies. Note that this implies different applications may actually experience a different MTBF on the same machine! Thus, we use three applications that present different conditions.

To gauge the effects of temperature control on MTBF and hence application performance, we formulate a model that relates total execution time of an application to reliability and the associated slowdown for temperature restraint. The model accounts for different execution speed at different frequencies, checkpointing overhead and recovery time depending on MTBF, and the additional overhead of experiencing and adapting to load imbalance. We validate the accuracy of our model for each application using a small experimental testbed.

We use our validated model to project the benefits of our scheme for larger machines. Our results point towards a tradeoff between improvement in reliability and the associated cost of applying temperature control. This tradeoff determines the optimal temperature threshold for a given application and machine size.

The following figure compares the machine efficiency (proportion of time spent doing useful work) for a 2D stencil application between a baseline run without temperature control and a constrained run with the temperature threshold set to 48C. Below 32K sockets, we get a lower efficiency than the baseline. However, above 32K sockets, our scheme starts outperforming the baseline case. For reference, the Blue Waters system at NCSA has nearly 50K sockets. For 256K sockets, our scheme is projected to operate the machine with an efficiency of 0.29 compared to 0.08 for the baseline. Finally, for 340K sockets, the baseline efficiency drops to 0.01, making the machine almost nonoperational, whereas our scheme can still operate the machine at an efficiency of 0.22.

exascale_race

These promising results encourage us to extend our work by investigating more detailed models, larger experimental systems, and more advanced fault tolerance protocols, such as message logging and parallel recovery.

This work is part of a research theme in our group: using adaptive runtime control to deal with the challenges presented by sophisticated applications and complexities of hardware. The Parallel Programming Laboratory has developed Charm++ for the past 20 years as a production quality parallel programming language, used in many CSE applications, including the Gordon Bell-winning biomolecular simulation program NAMD.

Please come attend our presentation at 1:30 pm on Tuesday in room 401, and other Charm++-related events.

Author Biography:

Osman Sarood is a final year PhD student in the UIUC Computer Science department. His research is focused on performance optimization under thermal and power constraints.

Esteban Meneses is a Research Assistant Professor working in the Center for Simulation and Modeling at the University of Pittsburgh. His research is focused on load balancing and fault tolerance techniques for large-scale parallel applications. He holds a PhD degree in Computer Science from UIUC.

Laxmikant Kale received his PhD in computer science from State University of New York, Stony Brook, in 1985. He joined the Computer Science faculty of UIUC as an Assistant Professor in 1985, where he is currently employed as a full Professor. His research spans parallel computing, including parallel programming abstractions, scalability, automatic load balancing, communication optimizations, and fault tolerance. He has collaboratively developed several scalable CSE applications.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industy updates delivered to you every week!

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together about 30 participants from industry, government and academia t Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Researchers Scale COSMO Climate Code to 4888 GPUs on Piz Daint

October 17, 2017

Effective global climate simulation, sorely needed to anticipate and cope with global warming, has long been computationally challenging. Two of the major obstacles are the needed resolution and prolonged time to compute Read more…

By John Russell

HPE Extreme Performance Solutions

Transforming Genomic Analytics with HPC-Accelerated Insights

Advancements in the field of genomics are revolutionizing our understanding of human biology, rapidly accelerating the discovery and treatment of genetic diseases, and dramatically improving human health. Read more…

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Cluster Competition coverage has come to its natural home: H Read more…

By Dan Olds

Data Vortex Users Contemplate the Future of Supercomputing

October 19, 2017

Last month (Sept. 11-12), HPC networking company Data Vortex held its inaugural users group at Pacific Northwest National Laboratory (PNNL) bringing together ab Read more…

By Tiffany Trader

AI Self-Training Goes Forward at Google DeepMind

October 19, 2017

DeepMind, Google’s AI research organization, announced today in a blog that AlphaGo Zero, the latest evolution of AlphaGo (the first computer program to defeat a Go world champion) trained itself within three days to play Go at a superhuman level (i.e., better than any human) – and to beat the old version of AlphaGo – without leveraging human expertise, data or training. Read more…

By Doug Black

Student Cluster Competition Coverage New Home

October 16, 2017

Hello computer sports fans! This is the first of many (many!) articles covering the world-wide phenomenon of Student Cluster Competitions. Finally, the Student Read more…

By Dan Olds

Intel Delivers 17-Qubit Quantum Chip to European Research Partner

October 10, 2017

On Tuesday, Intel delivered a 17-qubit superconducting test chip to research partner QuTech, the quantum research institute of Delft University of Technology (TU Delft) in the Netherlands. The announcement marks a major milestone in the 10-year, $50-million collaborative relationship with TU Delft and TNO, the Dutch Organization for Applied Research, to accelerate advancements in quantum computing. Read more…

By Tiffany Trader

Fujitsu Tapped to Build 37-Petaflops ABCI System for AIST

October 10, 2017

Fujitsu announced today it will build the long-planned AI Bridging Cloud Infrastructure (ABCI) which is set to become the fastest supercomputer system in Japan Read more…

By John Russell

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Intel Debuts Programmable Acceleration Card

October 5, 2017

With a view toward supporting complex, data-intensive applications, such as AI inference, video streaming analytics, database acceleration and genomics, Intel i Read more…

By Doug Black

Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

June 29, 2017

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms Read more…

By James Reinders

NERSC Scales Scientific Deep Learning to 15 Petaflops

August 28, 2017

A collaborative effort between Intel, NERSC and Stanford has delivered the first 15-petaflops deep learning software running on HPC platforms and is, according Read more…

By Rob Farber

How ‘Knights Mill’ Gets Its Deep Learning Flops

June 22, 2017

Intel, the subject of much speculation regarding the delayed, rewritten or potentially canceled “Aurora” contract (the Argonne Lab part of the CORAL “ Read more…

By Tiffany Trader

Oracle Layoffs Reportedly Hit SPARC and Solaris Hard

September 7, 2017

Oracle’s latest layoffs have many wondering if this is the end of the line for the SPARC processor and Solaris OS development. As reported by multiple sources Read more…

By John Russell

US Coalesces Plans for First Exascale Supercomputer: Aurora in 2021

September 27, 2017

At the Advanced Scientific Computing Advisory Committee (ASCAC) meeting, in Arlington, Va., yesterday (Sept. 26), it was revealed that the "Aurora" supercompute Read more…

By Tiffany Trader

Google Releases Deeplearn.js to Further Democratize Machine Learning

August 17, 2017

Spreading the use of machine learning tools is one of the goals of Google’s PAIR (People + AI Research) initiative, which was introduced in early July. Last w Read more…

By John Russell

GlobalFoundries Puts Wind in AMD’s Sails with 12nm FinFET

September 24, 2017

From its annual tech conference last week (Sept. 20), where GlobalFoundries welcomed more than 600 semiconductor professionals (reaching the Santa Clara venue Read more…

By Tiffany Trader

Graphcore Readies Launch of 16nm Colossus-IPU Chip

July 20, 2017

A second $30 million funding round for U.K. AI chip developer Graphcore sets up the company to go to market with its “intelligent processing unit” (IPU) in Read more…

By Tiffany Trader

Leading Solution Providers

Nvidia Responds to Google TPU Benchmarking

April 10, 2017

Nvidia highlights strengths of its newest GPU silicon in response to Google's report on the performance and energy advantages of its custom tensor processor. Read more…

By Tiffany Trader

Amazon Debuts New AMD-based GPU Instances for Graphics Acceleration

September 12, 2017

Last week Amazon Web Services (AWS) streaming service, AppStream 2.0, introduced a new GPU instance called Graphics Design intended to accelerate graphics. The Read more…

By John Russell

EU Funds 20 Million Euro ARM+FPGA Exascale Project

September 7, 2017

At the Barcelona Supercomputer Centre on Wednesday (Sept. 6), 16 partners gathered to launch the EuroEXA project, which invests €20 million over three-and-a-half years into exascale-focused research and development. Led by the Horizon 2020 program, EuroEXA picks up the banner of a triad of partner projects — ExaNeSt, EcoScale and ExaNoDe — building on their work... Read more…

By Tiffany Trader

Delays, Smoke, Records & Markets – A Candid Conversation with Cray CEO Peter Ungaro

October 5, 2017

Earlier this month, Tom Tabor, publisher of HPCwire and I had a very personal conversation with Cray CEO Peter Ungaro. Cray has been on something of a Cinderell Read more…

By Tiffany Trader & Tom Tabor

Cray Moves to Acquire the Seagate ClusterStor Line

July 28, 2017

This week Cray announced that it is picking up Seagate's ClusterStor HPC storage array business for an undisclosed sum. "In short we're effectively transitioning the bulk of the ClusterStor product line to Cray," said CEO Peter Ungaro. Read more…

By Tiffany Trader

Intel Launches Software Tools to Ease FPGA Programming

September 5, 2017

Field Programmable Gate Arrays (FPGAs) have a reputation for being difficult to program, requiring expertise in specialty languages, like Verilog or VHDL. Easin Read more…

By Tiffany Trader

IBM Advances Web-based Quantum Programming

September 5, 2017

IBM Research is pairing its Jupyter-based Data Science Experience notebook environment with its cloud-based quantum computer, IBM Q, in hopes of encouraging a new class of entrepreneurial user to solve intractable problems that even exceed the capabilities of the best AI systems. Read more…

By Alex Woodie

HPC Chips – A Veritable Smorgasbord?

October 10, 2017

For the first time since AMD's ill-fated launch of Bulldozer the answer to the question, 'Which CPU will be in my next HPC system?' doesn't have to be 'Whichever variety of Intel Xeon E5 they are selling when we procure'. Read more…

By Dairsie Latimer

  • arrow
  • Click Here for More Headlines
  • arrow
Share This