Wednesday, August 23, 2017

Session 3 Paper 2: Quantitative Network Monitoring with NetQRE (Yifei, Dong, Ankit, Sajal, Rajeev, Boon)

Presenter: Yifei Yuan
Authors: Yifei Yuan, Dong Lin, Ankit Mishra, Sajal Marwaha, Rajeev Alur, Boon Thau Loo

Dynamic updates are important to today's network management because both of traffic engineering and security events require timely response. To achieve this, quantitative network monitoring(including monitoring metrics and computation of values) is required. However, today's tools suffer from some limitations, such as focusing on low-level measurements, only providing ad-hoc solutions, or requiring significant programming expertise from network operators.

Yifei Yuan presented NetQRE, a language which can provide high-level abstractions. In the NetQRE language, expressions in stream functions are based on quantitative regular expressions(QRE). He also presented the compiler for NetQRE, which generates efficient NetQRE implementations automatically. In order to run with low memory footprint, they address challenges for evaluating each incoming packet without storing the history.

Yifei Yuan also presented several use cases including:
1. Flow-level traffic measurements
2. TCP state monitoring
3. application-level monitoring

Yuan et al. evaluated several aspects of the NetQRE prototype including the expressiveness of the language, performance in terms of throughput and memory footprint, and its capability for real-time monitoring setting. By comparing NetQRE with other tools, it is shown that NetQRE can effectively express a wide range of quantitative monitoring applications. It outperforms equivalent implements in other tools. And it can also support timely monitoring requirements.

You can check out this paper here.

Q&A:
Q: Can you compare your paper with the previous paper(Language-Directed Hardware Design for
Network Performance Monitoring)?
A: The Marple language uses key-value stores, but they still keep packet states. However in our case, we keep states across a sequence of packets. This is the benefit of using quantitative expressions instead of key-value.