MiAccLib







 

MIMOS Accelerator Library (Mi-AccLib)

The digital data explosion has exceeded petabytes and entered the zettabyte era, resulting in a big data challenge. MIMOS Mi-AccLib offers parallelised data analysis to enable ultra-speed big data processing across a heterogeneous platform.

Overview

MIMOS Mi-AccLib is an accelerator library designed to enable ultra-speed big data processing with data and process parallelism. This MIMOS proprietary software library offers a heterogeneous hardware accelerated computing environment to leverage hardware acceleration through parallelisation. The library capitalises on different processor capabilities while maintaining application usage needs to parallelise data analysis.

Features

Mi-AccLib comprises the following features:

  • Text/String Analytics
    This text/string library encompasses functions such as data matching, sorting, transformation, bitmapping, table operations (single/multiple columns key), date-searching and data cleansing. An application example is database data matching and cleansing using edit distance algorithm.

  • Financial Computation Algorithms
    Financial applications can also be compute-intensive. Mi-AccLib offers optimisation on calculations such as Value-At-Risk (VAR), data aggregation, conversion, equity, interest-rate yield factor and Monte Carlo.

  • Generic Parallelised Library
    The generic library provides generic APIs used for common functions such as sort and transform. It also includes specialised interfaces for security such as encryption/decryption functionalities, and various search/sort/transform algorithms.

  • Future Enhancements
    A Parallel In-Memory Database library will offer the IMDB GPU platform to users with improved common operations for rational databases. For the image and video library, customised intrusion and motion detection features with functions such as dilation and erosion filters, connected component labelling and background subtraction are to be included.

Technology Benefits

The main impacts of Mi-AccLib are:

  • Ultra-Speed Data Processing
    Mi-AccLib covers several areas for compute-intensive data such as financial and text/string operations and allows for users to directly use Mi-AccLib APIs for specific application needs, which run on multi-core CPU or GPGPU.

  • Transparent Heterogeneous Hardware Support
    Mi-AccLib APIs are designed to be processor agnostic and can run across different hardware platforms with ultra-speed processing capabilities but still guarantees the reliability of the data.

MIMOS Mi-AccLib system architecture

Performance Benchmark Results

System Requirements

Performance Benchmark Results

Patents

  • Load balancing for graphics processing units in an image processing system – Patent Pending
  • Scalable method and apparatus for parallel computation using hybrid heterogeneous processor capable of dynamic resource configuration – Patent Pending
  • Method and apparatus for accelerated SPARQL query processing – Patent Pending

Performance Benchmark Results

Performance Benchmark Results