high performance computing on graphics processing units: hgpu.org

We describe a novel deep packet inspection technique that provides precise quantitative measures for detecting data exfiltration. We point out the fundamental differences between our data leak detection and the conventional intrusion detection systems (IDS). The key to our solution is a powerful sampling algorithm and a sophisticated local alignment algorithm. Our sampling method has an unique comparable property – preserving the similarity of two input sequences during sampling. We have paralleled our new dynamic programming prototype on general-purpose graphics processing units to accelerate our detection system. We have extensively evaluated the scalability and security of our detection against several large datasets under real world data leak scenarios. Our algorithmic contributions are useful beyond the specific data leak detection problem.