2023 HPCC Systems Community Summit: Bitcoin Blockchain Parser + Optimization of Learning Trees...

Подробнее о видео

Building a Scalable and Efficient Bitcoin Blockchain Parser on the HPCC Systems Platform + Optimization of Learning Trees using ECL on HPCC Systems - Jyothi Shetty, Manonmani S & Sudershan K S, RV College of Engineering

This joint session includes two presentations from RV College of Engineering featuring the work from academic collaboration as part of the HPCC Systems Centre of Excellence (CoE) in Cognitive Intelligent Systems for Sustainable Solutions (CISSS).

Building a Scalable and Efficient Bitcoin Blockchain Parser on the HPCC Systems Platform, Jyothi Shetty

Many commercial applications are interested in analysing Bitcoin blockchain data, but due to the large volume of transactions the processing of transaction data becomes a challenge. A distributed and parallelized architecture such as HPCC Systems can be a solution to efficiently analyse such massive amounts of data. The work proposes to build a C++-based Bitcoin parsing algorithm embedded within ECL to create a robust and efficient blockchain parser. The proposed approach takes raw blockchain data in the form of blk.dat files as input and processes the headers and transaction data within each block. Furthermore, it maps input addresses to output addresses of previous transactions, constructing a connected network of blocks that accurately resembles the blockchain. By iteratively traversing this chain of blocks, the parsed contents are then written to a CSV file. The suggested implementation offers flexibility as it does not rely on external library dependencies and can be executed as a single ECL file. This is a work in progress study. An initial functional implementation of the parser code has been deployed successfully on hThor and the next step is to expand its capability to Thor so the parser can fully take advantage of HPCC Systems distributed and parallel processing capabilities. As the size of the blockchain data continues to grow, having a scalable and reliable ECL Bitcoin parser that enables retrieval of blockchain data with minimal overhead is quite beneficial. 

Optimization of Learning Trees using ECL on HPCC Systems, Manonmani S & Sudershan KS

Some core ECL functions on HPCC Systems, like the ones leveraged in the LearningTrees Machine Learning Bundle are recursive in nature and hence tend to demand higher computational effort under certain conditions. This talk presents an alternative solution for optimized execution of the LearningTrees algorithms in ECL via embedding python libraries. The proposed approach was tested on standard datasets and a significant decrease in execution time was observed with no impact on the accuracy of the models: while the embedding technique provided approximately the same accuracy levels as the current existing ML library functions, the execution time was reduced by 4-5 times making it an extremely fast and efficient technique. The target audience who will benefit from this talk includes industry professionals who wish to explore the resources for solving big data problems, technology enthusiasts willing to learn latest trends and technologies, and researchers working in big data and high-performance computing domains.

© 2023 LexisNexis Risk Solutions