Research | Kalyan Perumalla

Graph Networks

Fri, 01 Jul 2022 00:00:00 +0000

Parallel/Distributed implementation (1000s of GPUs), Graph algorithms, Graph Generation on multiple GPUs, Community detection-based solutions

Overview

See all items tagged Graph

Very large network generation (trillion-edge)
Very fast network generation
- GPU-based scalable algorithms
- Billion of edges per second
Scale-free and distribution-conformant networks
Community detection-based applications
- Electric grid distribution network inference

Fast GPU-Based Generation of Large Graph Networks From Degree Distributions

A novel algorithm has been designed, developed, and implemented on modern GPU accelerators, and benchmarked on networks with billions of edges, including Facebook and Twitter networks. Rates of generation exceed 50 billion edges per second.

Maksudul Alam, Kalyan Perumalla

PDF Cite DOI

Scale-Free Graph Networks with Trillions of Edges: Rapid Generation using 1000 GPUs

Our algorithm cuPPA generates scale-free networks using the preferential-attachment model, custom-designed to exploit multiple GPUs. We generate extremely large scale-free networks of 4 trillion edges in less than 8 minutes using 1,008 NVIDIA Volta GPUs of the Summit supercomputer. This represents the first ever graph network generation at this scale of parallel execution with over thousand GPUs. Moreover, our algorithm is uniquely suitable for generating networks in a streaming mode without the need for explicitly storing (writing to disk) the entire network, and is suitable for targeting even larger scales with quadrillions of edges.

Maksudul Alam, Kalyan Perumalla

PDF

On the Robustness of Network Community Structure Under Addition of Edges

We study the impact of edge additions on the community structure using Lancichinetti-Fortunato-Radicchi (LFR) benchmark networks. We show that, for a fixed network size, the impact of edge additions is greater on networks with initially weak community structure than on networks with strongly clustered structures. Also, we find that the perception of the impact is also dependent on the community detection algorithm used to uncover communities.

Pablo Moriano, Kalyan Perumalla

PDF DOI

Novel Parallel Algorithms for Fast Multi-GPU-Based Generation of Massive Scale-Free Networks

A novel parallel algorithm, cuPPA, is presented for generating random scale-free networks using the preferential attachment model. The algorithm is custom-designed for ‘single instruction multiple data (SIMD)’ for GPUs. Our algorithm is the first to exploit GPUs, and also the fastest implementation available today, for scale-free networks. On an NVidia GeForce 1080 GPU, cuPPA generates a scale-free network of two billion edges in less than 3 s. On multi-GPU platforms, cuPPA-Hash generates a scale-free network of 16 billion edges in less than 7 s using a machine consisting of 4 NVidia Tesla P100 GPUs.

Maksudul Alam, Kalyan Perumalla, Peter Sanders

PDF DOI

GPU-based parallel algorithm for generating massive scale-free networks using the preferential attachment model

A novel parallel algorithm, cuPPA, is presented for generating random scale-free networks using the preferential-attachment model. In one of the best cases, when executed on an NVidia GeForce 1080 GPU, cuPPA generates a scale-free network of two billion edges in less than 3 seconds.

Maksudul Alam, Kalyan Perumalla

PDF Cite Slides DOI

Generating Billion-Edge Scale-Free Networks in Seconds: Performance Study of a Novel GPU-based Preferential Attachment Model

A novel parallel algorithm is presented for generating random scale-free networks using the preferential-attachment model. In one of the best cases, when executed on an NVidia GeForce 1080 GPU, cuPPA generates a scale free network of a billion edges in less than 2 seconds.

Kalyan Perumalla, Maksudul Alam

Cite DOI

GPU-based Real-Time Execution of Vehicular Mobility Models in Large-Scale Road Network Scenarios

A methodology and its associated algorithms are presented for mapping a novel, field-based vehicular…

Kalyan Perumalla, Brandon Aaby, Srikanth Yoginath, Sudip Seal

PDF Cite

Network Simulation

A detailed introduction to the design, implementation, and use of network simulation tools is presen…

Richard Fujimoto, Kalyan Perumalla, George Riley

PDF Cite

Large-Scale Network Simulation - How Big? How Fast?

[Pub 44]

Richard Fujimoto, Kalyan Perumalla, Alfred Park, Hao Wu, Mostafa Ammar, George Riley

PDF Cite

AI ML Research

Wed, 01 Dec 2021 00:00:00 +0000

My research interests in AI/ML have focused on applying the technologies and software to applications in cyber-physical systems, digital twins, cybersecurity verification, and intelligent mechanical design. My research has largely used existing AI/ML methods and using them effectively in novel ways to solve difficult problems in various domains. Nevertheless, I am interested in the fundamental concepts underlying the methods and developing computational implementations from scratch, as, for example, in new Recurrent Neural Network implementations from scratch, not entirely relying on popular AI/ML packages.

From the computational point of view, my other major interests are in the end-to-end scalability of AI/ML solutions, especially in relation to novel hardware architectures, down-to-the-metal considerations, hardware-software interplay and codesign, and holistic scalability of workflows. Of particular interest are new AI/ML problems that are characterized by volumes and velocities massively higher than what typical desktop-based and cluster-based solutions address.

While reaping the benefits of well known and tested methods, my inclination is to keep an eye towards basic insights, unique combinations, and new techniques.

Overview

Projects

Deep CYBERIA
- CNN: Sensor detection
- RNN: Sensor time-series
CYVET
- NLP: CyBERT over BERT
- NLP: Claim detection
- CNN: Document classification
- CNN: Product detection
Digital Twin Framework
- RNN: Anomaly detection
DeepEx
- CNN: Material image classification
- VAE: Molecular Dynamics
Intelligent Design
- RL: Steering
Energy Management
- RL: Dynamic learning and adaptation
Kensor
- Critical Facility Monitoring
- Novel Markov modeling

Software

Python
Jupyter
numpy/no-numpy
CUDA
CUDNN
MPI
Tensorflow (2/Keras)
PyTorch

Organization

Sponsors: Various, across multiple projects
- US Department of Energy (DOE)
- US Department of Defense (DoD)
- Oak Ridge National Laboratory-Directed Research and Development (LDRD)
- Industry

Publications

Listed under the AI/ML category under Publications At-a-glance

Advanced Scientific Computing Research

Fri, 07 Mar 2025 00:00:00 +0000

A snapshot of Kalyan Perumalla’s Program Manager roles in the Advanced Scientific Computing Research in the Office of Science, U.S. Department of Energy (March 1, 2025).

Research | Kalyan Perumalla

Graph Networks

Overview

Related Publications

AI ML Research

Overview

Projects

Software

Organization

Publications

Advanced Scientific Computing Research

Program Management Areas