Search NASA⌕ Search

Engineering topics

Balaji, Pavan

Publications and source records attributed to Balaji, Pavan.

At least 19 records

MPICH

MPICH is a high-performance and widely portable implementation of the MPI-4 standard from the Argonne National Laboratory. This release has all MPI 4 functions and features required by the standard with the exception of support for the user-defined data representations for I/O.

ECP↗

RMACXX: An Efficient High-Level C++ Interface over MPI-3 RMA

Parallel scientific applications can benefit from de- coupling communication and synchronization. One-sided pro- gramming abstractions, which separate communication from syn- chronization, have in fact served as a motivation for partitioned global address space (PGAS) models. However, the use of PGAS models in application codes in a manner that fully exploits the benefit of these programming models requires significant development effort. Meanwhile, a vast majority of scientific codes already use the Message Passing Interface (MPI) and need convenient features to support application-specific one-sided communication scenarios. MPI Remote Memory Access (RMA) can be employed for this purpose. MPI is a low-level API, however, and developing applications with MPI RMA requires programmers to be well versed in its nuances. We present RMACXX, a compact set of C++ bindings to MPI-3 RMA, to ease the use of MPI RMA. Unlike other PGAS models, which may have interoperability issues with MPI, RMACXX is written on top of MPI and uses the same runtime as MPI. The basic functionality of RMACXX adds only a relatively small number of extra instructions (about 20) to the critical communication path. Moreover, RMACXX provides an intuitive API for building a wide variety of scientific applications while enjoying performance matching handwritten MPI-3 RMA codes.

MPI-3 RMA, one-sided communication, PGAS, C++, exp↗

Translational research in the MPICH project

The MPICH project is an example of translational research in computer science before that term was well known or even coined. The project began in 1992 as an effort to develop a portable, high-performance implementation of the emerging Message-Passing Interface (MPI) Standard. It has enabled the widespread adoption of MPI as a way to write scalable parallel applications on systems of all sizes including upcoming exascale supercomputers. In this paper, we describe how the translational research process was used in MPICH, how that led to its success, the challenges encountered and lessons learned, and how the process could be applied to other similar projects.

97 MATHEMATICS AND COMPUTING↗

Logically Parallel Communication for Fast MPI+Threads Applications

Supercomputing applications are increasingly adopting the MPI+threads programming model over the traditional "MPI everywhere" approach to better handle the disproportionate increase in the number of cores compared with other on-node resources. In practice, however, most applications observe a slower performance with MPI+threads primarily because of poor communication performance. Recent research efforts on MPI libraries address this bottleneck by mapping logically parallel communication, that is, operations that are not subject to MPI's ordering constraints to the underlying network parallelism. Domain scientists, however, typically do not expose such communication independence information because the existing MPI-3.1 standard's semantics can be limiting. Researchers had initially proposed user-visible endpoints to combat this issue, but such a solution requires intrusive changes to the standard (new APIs). The upcoming MPI-4.0 standard, on the other hand, allows applications to relax unneeded semantics and provides them with many opportunities to express logical communication parallelism. Here in this article, we show how MPI+threads applications can achieve high performance with logically parallel communication. Through application case studies, we compare the capabilities of the new MPI-4.0 standard with those of the existing one and user-visible endpoints (upper bound). Logical communication parallelism can boost the overall performance of an application by over 2x.

97 MATHEMATICS AND COMPUTING↗

Analysis of Threading Libraries for High Performance Computing

With the appearance of multi-/many core machines, applications and runtime systems have evolved in order to exploit the new on-node concurrency brought by new software paradigms. POSIX threads (Pthreads) was widely-adopted for that purpose and it remains as the most used threading solution in current hardware. Lightweight thread (LWT) libraries emerged as an alternative offering lighter mechanisms to tackle the massive concurrency of current hardware. In this article, we analyze in detail the most representative threading libraries including Pthread- and LWT-based solutions. In addition, to examine the suitability of LWTs for different use cases, we develop a set of microbenchmarks consisting of OpenMP patterns commonly found in current parallel codes, and we compare the results using threading libraries and OpenMP implementations. Moreover, we study the semantics offered by threading libraries in order to expose the similarities among different LWT application programming interfaces and their advantages over Pthreads. This article exposes that LWT libraries outperform solutions based on operating system threads when tasks and nested parallelism are required.

GLT↗

Analyzing the Performance Trade-Off in Implementing User-Level Threads

User-level threads have been widely adopted as a means of achieving lightweight concurrent execution without the costs of OS-level threads. Nevertheless, the costs of managing user-level threads represent a performance barrier that dictates how fine grained the concurrency exposed by an application can be without incurring significant overheads; this in turn may translate into insufficient parallelism to exploit highly parallel systems. This article is a deep dive into the fundamental costs in implementing user-level threads. We first identify that one of the highest sources of fork-join overheads stems from deviations, events that incur context switching during the execution of a thread and disrupt a run-to-completion execution. We then conduct an in-depth investigation of a wide spectrum of methods with respect to how they handle deviations while covering both parent- and child-first scheduling policies. Our methodology involves a comprehensive instruction- and cache-level analysis of all methods on several modern CPU architectures. Finally, the primary finding of our evaluation is that dynamic promotion methods that assume the absence of deviation and dynamically provide context-switching support offer the best trade-off between performance and capability when the likelihood of deviation is low.

97 MATHEMATICS AND COMPUTING↗

Counting Kmers for Biological Sequences at Large Scale

Counting the abundance of all the distinct kmers in biological sequence data is a fundamental step in bioinformatics. These applications include de novo genome assembly, error correction, etc. With the development of sequencing technology, the sequence data in a single project can reach Petabyte-scale or Terabyte-scale nucleotides. Counting demand for the abundance of these sequencing data is beyond the memory and computing capacity of single computing node, and how to process it efficiently is a challenge on a high-performance computing cluster. As such, we propose SWAPCounter, a highly scalable distributed approach for kmer counting. This approach is embedded with an MPI streaming I/O module for loading huge data set at high speed, and a counting bloom filter module for both memory and communication efficiency. By overlapping all the counting steps, SWAPCounter achieves high scalability with high parallel efficiency. In this work, the experimental results indicate that SWAPCounter has competitive performance with two other tools on shared memory environment, KMC 2 , and MSPKmerCounter. Moreover, SWAPCounter also shows the highest scalability under strong scaling experiments. In our experiment on Cetus supercomputer, SWAPCounter scales to 32,768 cores with 79% parallel efficiency (using 2048 cores as baseline) when processing 4 TB sequence data of 1000 Genomes. The source code of SWAPCounter is publicly available at https://github.com/mengjintao/SWAPCounter.

59 BASIC BIOLOGICAL SCIENCES↗

Scalable Deep Learning via I/O Analysis and Optimization

Scalable deep neural network training has been gaining prominence because of the increasing importance of deep learning in a multitude of scientific and commercial domains. Consequently, a number of researchers have investigated techniques to optimize deep learning systems. Much of the prior work has focused on runtime and algorithmic enhancements to optimize the computation and communication. Despite these enhancements, however, deep learning systems still suffer from scalability limitations, particularly with respect to data I/O. This situation is especially true for training models where the computation can be effectively parallelized, leaving I/O as the major bottleneck. In fact, our analysis shows that I/O can take up to 90% of the total training time. Thus, in this article, we first analyze LMDB, the most widely used I/O subsystem of deep learning frameworks, to understand the causes of this I/O inefficiency. Based on our analysis, we propose LMDBIO—an optimized I/O plugin for scalable deep learning. LMDBIO includes six novel optimizations that together address the various shortcomings in existing I/O for deep learning. Lastly, our experimental results show that LMDBIO significantly outperforms LMDB in all cases and improves overall application performance by up to 65-fold on a 9,216-core system.

97 MATHEMATICS AND COMPUTING↗

Dynamic Adaptable Asynchronous Progress Model for MPI RMA Multiphase Applications

Casper is a process-based asynchronous progress model for MPI one-sided communication on multi-and many-core architectures. The one-sided communication is not truly one-sided in most MPI implementations: the target process still relies on software progress to complete incoming operations. Casper allows the user to specify an arbitrary number of cores dedicated to background ghost processes and transparently redirects the RMA operations to ghost processes by utilizing the PMPI redirection and MPI-3 shared-memory technologies. Although Casper benefits applications that suffer from lack of asynchronous progress, the operation redirection design might not support complex multiphase applications effectively, which often involve dynamically changing communication density and computing workloads. In this paper, we present an adaptive mechanism in Casper to address the limitation of static asynchronous progress in multiphase applications. For this we exploit two adaptive strategies, a user-guided strategy and a fully transparent and automatic strategy based on self-profiling and prediction, to dynamically reconfigure the asynchronous progress in Casper according to real-time performance characteristics during multiphase execution. We evaluate the adaptive approaches in both microbenchmarks and a real quantum chemistry application suite, NWChem, on the Cray XC30 supercomputer and an Intel Omni-Path cluster.

97 MATHEMATICS AND COMPUTING↗

Argobots: A Lightweight Low-Level Threading and Tasking Framework

In the past few decades, a number of user-level threading and tasking models have been proposed in the literature to address the shortcomings of OS-level threads, primarily with respect to cost and flexibility. Current state-of-the-art user-level threading and tasking models, however, are either too specific to applications or architectures or are not as powerful or flexible. In this paper, we present Argobots, a lightweight, low-level threading and tasking framework that is designed as a portable and performant substrate for high-level programming models or runtime systems. Argobots offers a carefully designed execution model that balances generality of functionality with providing a rich set of controls to allow specialization by the user or high-level programming model. We describe the design, implementation, and optimization of Argobots and present integrations with three example high-level models: OpenMP, MPI, and co-located I/O service. Evaluations show that (1) Argobots outperforms existing generic threading runtimes; (2) our OpenMP runtime offers more efficient interoperability capabilities than production OpenMP runtimes do; (3) when MPI interoperates with Argobots instead of Pthreads, it enjoys reduced synchronization costs and better latency hiding capabilities; and (4) I/O service with Argobots reduces interference with co-located applications, achieving performance competitive with that of the Pthreads version.

Seo, Sangmin (ORCID:0000000295159836)↗

On the adequacy of lightweight thread approaches for high-level parallel programming models

High-level parallel programming models (PMs) are becoming crucial in order to extract the computational power of current on-node multi-threaded parallelism. The most popular PMs, such as OpenMP or OmpSs, are directive-based: the complexity of the hardware is hidden by the underlying runtime system, improving coding productivity. The implementations of OpenMP usually rely on POSIX threads (pthreads), offering excellent performance for coarse-grained parallelism and a perfect match with the current hardware. OmpSs is a task oriented PM based on an ad hoc runtime solution called Nanos++; it is the precursor of the tasking parallelism in the OpenMP tasking specification. A recent trend in runtimes and applications points to leveraging massive on-node parallelism in conjunction with fine-grained and dynamic scheduling paradigms. In this paper we analyze the behavior of the OpenMP and OmpSs PMs on top of the recently emerged Generic Lightweight Threads (GLT) API. GLT exposses a common API for lightweight thread (LWT) libraries that offers the possibility of running the same application over different native LWT solutions. We describe the design details of those high-level PMs implemented on top of GLT and analyze different scenarios in order to assess where the use of LWTs may benefit application performance. Our work reveals those scenarios where LWTs overperform pthread-based solutions and compares the performance between an ad hoc solution and a generic implementation.

97 MATHEMATICS AND COMPUTING↗

Memory Compression Techniques for Network Address Management in MPI

MPI allows applications to treat processes as a logical collection of integer ranks for each MPI communicator, while internally translating these logical ranks into actual network addresses. In current MPI implementations the management and lookup of such network addresses use memory sizes that are proportional to the number of processes in each communicator. In this paper, we propose a new mechanism, called AV-Rankmap, for managing such translation. AV-Rankmap takes advantage of logical patterns in rank-address mapping that most applications naturally tend to have, and it exploits the fact that some parts of network address structures are naturally more performance critical than others. It uses this information to compress the memory used for network address management. We demonstrate that AV-Rankmap can achieve performance similar to or better than that of other MPI implementations while using significantly less memory.

Guo, Yanfei↗

Towards Scalable Deep Learning via I/O Analysis and Optimization

Deep learning systems have been growing in prominence as a way to automatically characterize objects, trends, and anomalies. Given the importance of deep learning systems, researchers have been investigating techniques to optimize such systems. An area of particular interest has been using large supercomputing systems to quickly generate effective deep learning networks: a phase often referred to as “training” of the deep learning neural network. As we scale existing deep learning frameworks—such as Caffe—on these large supercomputing systems, we notice that the parallelism can help improve the computation tremendously, leaving data I/O as the major bottleneck limiting the overall system scalability. In this paper, we first present a detailed analysis of the performance bottlenecks of Caffe on large supercomputing systems. Our analysis shows that the I/O subsystem of Caffe—LMDB—relies on memory-mapped I/O to access its database, which can be highly inefficient on large-scale systems because of its interaction with the process scheduling system and the network-based parallel filesystem. Based on this analysis, we then present LMDBIO, our optimized I/O plugin for Caffe that takes into account the data access pattern of Caffe in order to vastly improve I/O performance. Our experimental results show that LMDBIO can improve the overall execution time of Caffe by nearly 20-fold in some cases.

Pumma, Sarunya↗