Celebrating advances in domain-specific languages
VMware is pleased to announce the 2018 recipient of the early career Systems Research Award: Tiark Rompf, Assistant Professor of Computer Science at Purdue University. Tiark is recognized for radically new approaches to performance- and safety-critical systems, in particular through rethinking the role and relationship between high-level and low-level languages. Professor Rompf’s university will receive a gift of US $100,000 in support of his highly-valued research on programming languages and compilers.
Professor Rompf’s systems-oriented approach is illustrated well by his far-ranging explorations of lightweight modular staging (LMS), a platform and methodology for enabling run-time code generation. LMS is particularly helpful as a foundation for practical domain-specific languages (DSLs), which have been an important theme in programming languages for over twenty years. In principle, DSLs enable subject-matter experts to write programs that are comprehensible, correct, and efficient; a great example is SQL, which had enormous impact on the database community. But designing and implementing a DSL remains challenging and expensive. Martin Odersky, the lead behind the Scala programming language (and also Professor Rompf’s Ph.D. advisor), explains that “Recently there have been two important breakthroughs in Domain-Specific Languages. The first was the notion of embedded DSLs, which allows the machinery of a higher-level language to be re-used as the foundation of a DSL. A second is the idea of staging, which allows an embedded DSL to be highly efficient through code generation. In LMS, Tiark has developed an elegant theory of staging, and he has realized that idea in ways that are relevant to real-world DSL and systems builders.”
Kunle Olukotun, an early collaborator with Tiark and Martin who leads the Stanford Pervasive Parallelism Lab, says “Our team originally explored Tiark’s work as a way to bring sophisticated compilation techniques to expressive, user-friendly DSLs. By incorporating LMS into the Delite compiler framework, we were able to bridge the gap between productive languages and high performance, parallel and heterogeneous hardware. Now, as we work to democratize machine learning, we find that LMS-based compilation frameworks like Delite and Spatial provide a seamless way to target emerging domain-specific hardware platforms (such as FPGA-accelerated CPUs) from high-level programs. LMS gives us efficiency without compromising expressivity.”
Recent works by Tiark’s group, Flare and LB2, have explored data processing and query optimization, leading to a compelling back-end for Spark that achieves orders of magnitude speedups by using LMS to compile high-level queries into lower-level code. Also timely are Tiark’s applications of LMS to machine learning, spanning his early work on OptiML and his recent work on Lantern, which proposes new directions in differentiable programming to enable systems that are both expressive (like PyTorch) and performant (like TensorFlow).
The VMware Systems Research Award celebrates early-career faculty within the first five years of their first tenure-track appointment. Tiark, observed VMware Fellow Pratap Subrahmanyam, “undertook work in a challenging area, pursued it for a number of years, and generated what is emerging as a fundamental and important idea with significant current impact and future potential. That he has done so in a way that is both principled and practical is what makes him a role-modeling systems researcher in our eyes”. Says Ole Agesen, also a VMware fellow, “This year alone, Tiark had publications in a remarkable breadth of top-tier academic venues in multiple application domains (SIGMOD, NIPS), and in a range of systems and theory conferences (OSDI and POPL); plus his work is being used and explored in several startups and established industry companies. These are impressive indicators of the holistic thinking and broad-ranging impact we look for when placing this award.”
This year, the selection committee was chaired by Professor Mike Stonebraker (MIT) and included Ole Agesen (VMware), Professor Edouard Bugnion (EPFL), Professor Greg Ganger (CMU), Chris Ramming (VMware), Professor Jennifer Rexford (Princeton), and Pratap Subrahmanyam (VMware). VMware sees the award as one way to support and give back to the academic research community, which plays a crucial role in exploring new technology.
–Chris Ramming, Senior Director, VMware Research & Innovation