Publications
Driving big data with big compute
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
Big Data (as embodied by Hadoop clusters) and Big Compute (as embodied by MPI clusters) provide unique capabilities for storing and processing large volumes of data. Hadoop clusters make distributed computing readily accessible to the Java community and MPI clusters provide high parallel efficiency for compute intensive workloads. Bringing the big data and big compute communities together is an active area of research. The LLGrid team has developed and deployed a number of technologies that aim to provide the best of both worlds. LLGrid MapReduce allows the map/reduce parallel programming model to be used quickly and efficiently in any language on any compute cluster. D4M (Dynamic Distributed Dimensional Data Model) provided a high level distributed arrays interface to the Apache Accumulo database. The accessibility of these technologies is assessed by measuring the effort to use these tools and is typically a few lines of code. The performance is assessed by measuring the insert rate into the Accumulo database. Using these tools a database insert rate of 4M inserts/second has been achieved on an 8 node cluster.
Summary
Big Data (as embodied by Hadoop clusters) and Big Compute (as embodied by MPI clusters) provide unique capabilities for storing and processing large volumes of data. Hadoop clusters make distributed computing readily accessible to the Java community and MPI clusters provide high parallel efficiency for compute intensive workloads. Bringing the...
READ MORE
Cluster-based 3D reconstruction of aerial video
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
Summary
Large-scale 3D scene reconstruction using Structure from Motion (SfM) continues to be very computationally challenging despite much active research in the area. We propose an efficient, scalable processing chain designed for cluster computing and suitable for use on aerial video. The sparse bundle adjustment step, which is iterative and difficult to parallelize, is accomplished by partitioning the input image set, generating independent point clouds in parallel, and then fusing the clouds and combining duplicate points. We compare this processing chain to a leading parallel SfM implementation, which exploits fine-grained parallelism in various matrix operations and is not designed to scale beyond a multi-core workstation with GPU. We show our cluster-based approach offers significant improvement in scalability and runtime while producing comparable point cloud density and more accurate point location estimates.
Summary
Large-scale 3D scene reconstruction using Structure from Motion (SfM) continues to be very computationally challenging despite much active research in the area. We propose an efficient, scalable processing chain designed for cluster computing and suitable for use on aerial video. The sparse bundle adjustment step, which is iterative and difficult...
READ MORE
Scalable cryptographic authentication for high performance computing
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
High performance computing (HPC) uses supercomputers and computing clusters to solve large computational problems. Frequently HPC resources are shared systems and access to restricted data sets or resources must be authenticated. These authentication needs can take multiple forms, both internal and external to the HPC cluster. A computational stack that uses web services among nodes in the HPC may need to perform authentication between nodes of the same job or a job may need to reach out to data sources outside the HPC. Traditional authentication mechanisms such as passwords or digital certificates encounter issues with the distributed and potentially disconnected nature of HPC systems. Distributing and storing plain-text passwords or cryptographic keys among nodes in a HPC system without special protection is a poor security practice. Systems that reach back to the user's terminal for access to the authenticator are possible, but only in fully interactive supercomputing where connectivity to the user's terminal can be guaranteed. Point solutions can be enabled for these use cases, such as software-based role or self-signed certificates, however they require significant expertise in digital certificates to configure. A more general solution is called for that is both secure and easy to use. This paper presents an overview of a solution implemented on the interactive, on-demand LLGrid computing system at MIT Lincoln Laboratory and its use to solve one such authentication problem.
Summary
High performance computing (HPC) uses supercomputers and computing clusters to solve large computational problems. Frequently HPC resources are shared systems and access to restricted data sets or resources must be authenticated. These authentication needs can take multiple forms, both internal and external to the HPC cluster. A computational stack that...
READ MORE
HPC-VMs: virtual machines in high performance computing systems
September 10, 2012
Journal Article
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
The concept of virtual machines dates back to the 1960s. Both IBM and MIT developed operating system features that enabled user and peripheral time sharing, the underpinnings of which were early virtual machines. Modern virtual machines present a translation layer of system devices between a guest operating system and the host operating system executing on a computer system, while isolating each of the guest operating systems from each other. In the past several years, enterprise computing has embraced virtual machines to deploy a wide variety of capabilities from business management systems to email server farms. Those who have adopted virtual deployment environments have capitalized on a variety of advantages including server consolidation, service migration, and higher service reliability. But they have also ended up with some challenges including a sacrifice in performance and more complex system management. Some of these advantages and challenges also apply to HPC in virtualized environments. In this paper, we analyze the effectiveness of using virtual machines in a high performance computing (HPC) environment. We propose adding some virtual machine capability to already robust HPC environments for specific scenarios where the productivity gained outweighs the performance lost for using virtual machines. Finally, we discuss an implementation of augmenting virtual machines into the software stack of a HPC cluster, and we analyze the affect on job launch time of this implementation.
Summary
The concept of virtual machines dates back to the 1960s. Both IBM and MIT developed operating system features that enabled user and peripheral time sharing, the underpinnings of which were early virtual machines. Modern virtual machines present a translation layer of system devices between a guest operating system and the...
READ MORE
Benchmarking parallel eigen decomposition for residuals analysis of very large graphs
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
Summary
Graph analysis is used in many domains, from the social sciences to physics and engineering. The computational driver for one important class of graph analysis algorithms is the computation of leading eigenvectors of matrix representations of a graph. This paper explores the computational implications of performing an eigen decomposition of a directed graph's symmetrized modularity matrix using commodity cluster hardware and freely available eigensolver software, for graphs with 1 million to 1 billion vertices, and 8 million to 8 billion edges. Working with graphs of these sizes, parallel eigensolvers are of particular interest. Our results suggest that graph analysis approaches based on eigen space analysis of graph residuals are feasible even for graphs of these sizes.
Summary
Graph analysis is used in many domains, from the social sciences to physics and engineering. The computational driver for one important class of graph analysis algorithms is the computation of leading eigenvectors of matrix representations of a graph. This paper explores the computational implications of performing an eigen decomposition of...
READ MORE
Scalable cryptographic authentication for high performance computing
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
High performance computing (HPC) uses supercomputers and computing clusters to solve large computational problems. Frequently HPC resources are shared systems and access to restricted data sets or resources must be authenticated. These authentication needs can take multiple forms, both internal and external to the HPC cluster. A computational stack that uses web services among nodes in the HPC may need to perform authentication between nodes of the same job or a job may need to reach out to data sources outside the HPC. Traditional authentication mechanisms such as passwords or digital certificates encounter issues with the distributed and potentially disconnected nature of HPC systems. Distributing and storing plain-text passwords or cryptographic keys among nodes in a HPC system without special protection is a poor security practice. Systems that reach back to the user's terminal for access to the authenticator are possible, but only in fully interactive supercomputing where connectivity to the user's terminal can be guaranteed. Point solutions can be enabled for these use cases, such as software-based role or self-signed certificates, however they require significant expertise in digital certificates to configure. A more general solution is called for that is both secure and easy to use. This paper presents an overview of a solution implemented on the interactive, on-demand LLGrid computing system at MIT Lincoln Laboratory and its use to solve one such authentication problem.
Summary
High performance computing (HPC) uses supercomputers and computing clusters to solve large computational problems. Frequently HPC resources are shared systems and access to restricted data sets or resources must be authenticated. These authentication needs can take multiple forms, both internal and external to the HPC cluster. A computational stack that...
READ MORE
Large scale network situational awareness via 3D gaming technology
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
Obtaining situational awareness of network activity across an enterprise presents unique visualization challenges. IT analysts are required to quickly gather and correlate large volumes of disparate data to identify the existence of anomalous behavior. This paper will show how the MIT Lincoln Laboratory LLGrid Team has approached obtaining network situational awareness utilizing the Unity 3D video game engine. We have developed a 3D environment of the physical plant in the format of a networked multi player First Person Shooter (FPS) to demonstrate a virtual depiction of the current state of the network and the machines operating on the network. Within the game or virtual world an analyst or player can gather critical information on all network assets as well as perform physical system actions on machines in question. 3D gaming technology provides tools to create an environment that is both visually familiar to the player as well display immense amounts of system data in a meaningful and easy to absorb format. Our prototype system was able to monitor and display 5000 assets in ~10% of the time of our network time window.
Summary
Obtaining situational awareness of network activity across an enterprise presents unique visualization challenges. IT analysts are required to quickly gather and correlate large volumes of disparate data to identify the existence of anomalous behavior. This paper will show how the MIT Lincoln Laboratory LLGrid Team has approached obtaining network situational...
READ MORE
Driving big data with big compute
September 10, 2012
Conference Paper
Published in:
HPEC 2012: IEEE Conf. on High Performance Extreme Computing, 10-12 September 2012.
Topic:
R&D area:
R&D group:
Summary
Big Data (as embodied by Hadoop clusters) and Big Compute (as embodied by MPI clusters) provide unique capabilities for storing and processing large volumes of data. Hadoop clusters make distributed computing readily accessible to the Java community and MPI clusters provide high parallel efficiency for compute intensive workloads. Bringing the big data and big compute communities together is an active area of research. The LLGrid team has developed and deployed a number of technologies that aim to provide the best of both worlds. LLGrid MapReduce allows the map/reduce parallel programming model to be used quickly and efficiently in any language on any compute cluster. D4M (Dynamic Distributed Dimensional Data Model) provided a high level distributed arrays interface to the Apache Accumulo database. The accessibility of these technologies is assessed by measuring the effort to use these tools and is typically a few lines of code. The performance is assessed by measuring the insert rate into the Accumulo database. Using these tools a database insert rate of 4M inserts/second has been achieved on an 8 node cluster.
Summary
Big Data (as embodied by Hadoop clusters) and Big Compute (as embodied by MPI clusters) provide unique capabilities for storing and processing large volumes of data. Hadoop clusters make distributed computing readily accessible to the Java community and MPI clusters provide high parallel efficiency for compute intensive workloads. Bringing the...
READ MORE
Query-by-example using speaker content graphs
September 9, 2012
Conference Paper
Published in:
INTERSPEECH 2012: 13th Annual Conf. of the Int. Speech Communication Assoc., 9-13 September 2012.
Topic:
R&D area:
Summary
We describe methods for constructing and using content graphs for query-by-example speaker recognition tasks within a large speech corpus. This goal is achieved as follows: First, we describe an algorithm for constructing speaker content graphs, where nodes represent speech signals and edges represent speaker similarity. Speech signal similarity can be based on any standard vector-based speaker comparison method, and the content graph can be constructed using an efficient incremental method for streaming data. Second, we apply random walk methods to the content graph to find matching examples to an unlabeled query set of speech signals. The content-graph based method is contrasted to a more traditional approach that uses supervised training and stack detectors. Performance is compared in terms of information retrieval measures and computational complexity. The new content-graph based method is shown to provide a promising low-complexity scalable alternative to standard speaker recognition methods.
Summary
We describe methods for constructing and using content graphs for query-by-example speaker recognition tasks within a large speech corpus. This goal is achieved as follows: First, we describe an algorithm for constructing speaker content graphs, where nodes represent speech signals and edges represent speaker similarity. Speech signal similarity can be...
READ MORE
Supervector LDA - a new approach to reduced-complexity i-vector language recognition
September 9, 2012
Conference Paper
Published in:
INTERSPEECH 2012: 13th Annual Conf. of the Int. Speech Communication Assoc., 9-13 September 2012.
Topic:
R&D area:
Summary
In this paper, we extend our previous analysis of Gaussian Mixture Model (GMM) subspace compensation techniques using Gaussian modeling in the supervector space combined with additive channel and observation noise. We show that under the modeling assumptions of a total-variability i-vector system, full Gaussian supervector scoring can also be performed cheaply in the total subspace, and that i-vector scoring can be viewed as an approximation to this. Next, we show that covariance matrix estimation in the i-vector space can be used to generate PCA estimates of supervector covariance matrices needed for Joint Factor Analysis (JFA). Finally, we derive a new technique for reduced-dimension i-vector extraction which we call Supervector LDA (SV-LDA), and demonstrate a 100-dimensional i-vector language recognition system with equivalent performance to a 600-dimensional version at much lower complexity.
Summary
In this paper, we extend our previous analysis of Gaussian Mixture Model (GMM) subspace compensation techniques using Gaussian modeling in the supervector space combined with additive channel and observation noise. We show that under the modeling assumptions of a total-variability i-vector system, full Gaussian supervector scoring can also be performed...
READ MORE