GPU-Based Algorithms

Overview

I also worked on other problems related to network mining and management. Mining and analyzing the network data requires high throughput, low latency data lookup. I designed and developed GRT (GPU-based Radix Tree), a new and highly efficient radix tree-based index-searching implementation for the GPU that supports both conventional point and range queries. GRT outperforms existing GPU-based index-searching system for a wide range of benchmarks. GRT achieves a high throughput of $106$ and $130$ million lookups per second for sparse and dense keys respectively, for point queries, with a large dataset of $64$ million $32$-bit keys.  It yields $600$ and $1000$ million lookups per second for sparse and dense keys for range queries.

Publications

A paper summarizing the work has been submitted to a top-tier conference.