Choice of Parallelism: Multi-GPU Driven Pipeline for Huge Academic Backbone Network

TL;DR

This paper introduces a multi-GPU pipeline for processing massive session data from the Japanese SINET network, significantly accelerating data discrimination and grouping tasks using GPU and a two-stage map-reduce design.

Contribution

It presents a novel multi-GPU-driven pipeline with a tiling pattern for efficient large-scale session data processing in an academic network context.

Findings

Processed 1.2 to 1.6 billion session streams within 24 hours.

Achieved acceleration of session data discrimination using GPU.

Implemented a two-stage map-reduce with CPU and GPU for efficient data grouping.

Abstract

Science Information Network (SINET) is a Japanese academic backbone network for more than 800 research institutions and universities. In this paper, we present a multi-GPU-driven pipeline for handling huge session data of SINET. Our pipeline consists of ELK stack, multi-GPU server, and Splunk. A multi-GPU server is responsible for two procedures: discrimination and histogramming. Discrimination is dividing session data into ingoing/outgoing with subnet mask calculation and network address matching. Histogramming is grouping ingoing/outgoing session data into bins with map-reduce. In our architecture, we use GPU for the acceleration of ingress/egress discrimination of session data. Also, we use a tiling design pattern for building a two-stage map-reduce of CPU and GPU. Our multi-GPU-driven pipeline has succeeded in processing huge workloads of about 1.2 to 1.6 billion session streams (500GB-650GB) within 24 hours.