(2003 - 2008)
My PhD dissertation work proposed an architecture for constructing new implementations of standard Internet protocols with integrated formal methods such as model checking and functional programming that were then not used in deployed servers.
In the late 90s while working at MVACS on the Mars Polar Lander, I found myself using the secure OpenBSD operating system to deploy the self-hosted service that Nick Ludlam and I have run together ever since. I became an OpenBSD developer with commit rights and went to several hackathons, a sample of which you can read in OpenBSD C2K5 thoughts. Back then, my primary open source experience was working on C code in the OpenBSD base system and in PHP code while hacking on the popular Horde/IMP groupware system for my own email.
I rapidly tired of hacking in C code and looked for safer alternatives. While procrastinating over PhD coffee with David Scott he suggested I look into writing a system daemon in OCaml. Why not have a go at a SSH server written entirely in a type-safe functional language? Being a PhD student sorely in need of a challenge, I took up the challenge.
There were a couple of different challenges involved:
All this work resulted in the Melange framework that I put together in OCaml and evaluated, and published in my “Creating High-Performance, Statically Type-Safe Network Applications” PhD thesis with the following abstract:
A typical Internet server finds itself in the middle of a virtual battleground, under constant threat from worms, viruses and other malware seeking to subvert the original intentions of the programmer. In particular, critical Internet servers such as OpenSSH, BIND and Sendmail have had numerous security issues ranging from low-level buffer overflows to subtle protocol logic errors. These problems have cost billions of dollars as the growth of the Internet exposes increasing numbers of computers to electronic malware. Despite the decades of research on techniques such as model-checking, type-safety and other forms of formal analysis, the vast majority of server implementations continue to be written unsafely and informally in C/C++.
In this dissertation we propose an architecture for constructing new implementations of standard Internet protocols which integrates mature formal methods not currently used in deployed servers: (i) static type systems from the ML family of functional languages; (ii) model checking to verify safety properties exhaustively about aspects of the servers; and (iii) generative meta-programming to express high-level constraints for the domain-specific tasks of packet parsing and constructing non-deterministic state ma- chines. Our architecture -— dubbed MELANGE -— is based on Objective Caml and contributes two domain-specific languages: (i) the Meta Packet Language (MPL), a data description language used to describe the wire format of a protocol and output statically type-safe code to handle network traffic using high-level functional data structures; and (ii) the Statecall Policy Language (SPL) for constructing non-deterministic finite state automata which are embedded into applications and dynamically enforced, or translated into PROMELA and statically model-checked. Our research emphasises the importance of delivering efficient, portable code which is feasible to deploy across the Internet. We implemented two complex protocols -— SSH and DNS -— to verify our claims, and our evaluation shows that they perform faster than their standard counterparts OpenSSH and BIND, in addition to providing static guarantees against some classes of errors that are currently a major source of security problems.
I didn't do much on this immediately after submitting my thesis since I was busy working on Xen Virtualisation from 2006-2009 or so. However, the first thing I did when I quit Citrix was to start the MirageOS project (the successor to Melange) with Thomas Gazagnaire and David Scott in order to develop better personal data infrastructure with Personal Containers. This formed the foundation for my subsequent research into library operating systems and the concept of Unikernels. Read more about the subsequent work there, or sample “Turning down the LAMP: Software Specialisation for the Cloud” to get a taster of the direction Melange evolved in.
Reflecting on my PhD research in 2021, I think that it was a pretty good piece of systems research. It didn't make any deep contributions to formal verification or programming language research, but it did posit a clear systems thesis and implement and evaluate it without a huge team being involved. That's more difficult to do these days in the era of large industrial research teams dominating the major conferences, but certainly not impossible.
Choosing a good topic for systems research is crucial, since the context you do the research in is as important as the results you come up with. Much of my subsequent career has been influenced by the "crazy challenge" that David Scott set me back in 2003 to do systems programming in a functional language, with all the intellectual and engineering challenges that came along with that extreme (back in 2003) position.