Research

Network Verification is an emerging field with startups like Forward Networks and company efforts like Amazon's Automated Verification group.  Existing work deals with verifying router configurations and ignores the Zone files set up manually for the Domain Name Service (DNS).   An example of our most recent work in this area that has found bugs in Microsoft and Akamai zone files is the SIGCOMM 2020 Best Student Paper Groot 

Most of my early career was focused on making the Internet faster with a special focus on algorithms for fast routers that can be implemented at Gigabit speeds. Two examples of algorithms that have been widely deployed are algorithms for scheduling packets at routers for fairness (DRR) and algorithms for longest matching prefix at routers  such as Binary Search on Prefix Lengths.

Realizing that the vision of Network Design Automation requires more than verification, we have been working at UCLA on a debugging assistant for networks that suggests debugging queries using machine learning on unstructured inputs such as free form text reports.

We wish to improve the reliability of networks by modelling them as mathematical objects and proving them correct.  Unfortunately,  as with many formal methods, network operators lack formal specifications.  Instead, we only attempt to find bugs. As an example, of how we do  this by network specific data mining, see the 2020 NSDI paper and tool (which is being used in the Microsoft wide area network) called  Self-Starter

My mid career work was in finding algorithms for finding insights in data arriving at high speeds using network specific sketches.  An example of my early work in this area is Sample-and-Hold and Multistage filters that were deployed in Cisco routers.  We later applied these ideas to detect attacks like worms in EarlyBird, forming a company called NetSift that was acquired by Cisco.

Unlike the ML project, in this project we are doing at UCLA, we ask ourselves what additional debugging primitives we could add to routers to make network problems  (e.g., route flapping, lost packets) easier to diagnose from symptoms.