PPar Lunch Series 2016/17

Date                  

Speaker

Title / Abstract

Performance, Portability and Productivity for Room Acoustics Simulations

As Moore’s Law loses relevance and the parallel computing landscape expands, it is becoming increasingly difficult for computational scientists to write performant portable simulations targeting new architectures without rewriting their codes. Many of the relevant advancements this domain are often tested out only on simple benchmarks, which can create gaps in the utility of new frameworks for more complicated scientific codes.  My research approaches this problem from the bottom up, by developing a solution specific to the application domain at hand: room acoustics. Room acoustics simulations model sound in 3D spaces and are useful in fields such as virtual reality and musical composition. In this talk I will discuss my work towards developing a performant, portable and productive room acoustics model using the Lift framework, which will be applicable for other wave-based simulations.

Safe, Expressive, and Effectful

This talk will summarize my recent work on integrating linear typing and functional programming, explore the relevance of linear typing to problems in domains such as concurrency and systems programming, and discuss the use of my approach to express (as libraries) other approaches to effectful programming, such as region types, ownership types, or adoption and focus.

Quantum Computing (some interesting bits and bobs)

During this talk I will try to explain some of the ways in which quantum computing is different from classical computing and some of the algorithms and ideas which best exhibit this separation. I will also end with some of the work from my masters dissertation which explored how we can test for the amount of quantum power a computer holds.

Controlling Collective Adaptive Systems

In this talk I will introduce a class of systems called collective adaptive systems, give some reasons why they are interesting, and talk about some of my currently planned work on gaining insight on how to effectively control such systems.

ePython: An implementation of Python for the many-core Epiphany co-processor

The Epiphany is a many-core, low power, low on-chip memory architecture and one can very cheaply gain access to a number of parallel cores which is beneficial for HPC education and prototyping. The very low power nature of these architectures also means that there is potential for their use in future HPC machines; however, there is a high barrier to entry in programming them due to the associated complexities and immaturity of supporting tools. I will talk about ePython, a subset of Python for the Epiphany and similar many-core co-processors. Due to the limited on-chip memory per core we have developed a new Python interpreter and this, combined with additional support for parallelism, has meant that novices can take advantage of Python to very quickly write parallel codes on the Epiphany and explore concepts of HPC using a smaller scale parallel machine. The high level nature of Python opens up new possibilities on the Epiphany and other similar architectures that the community have already started to adopt and use to explore concepts of parallelism and HPC.

Hardware Adaptive Simulation of Physical Deformation and Interaction Using GPUs

Finite Element Methods (FEM) are a well known technique to solve elasticity problems with their ability to provide a continuous description of the underlying physical equations. The associated advantages include the ability to accurately represent deformation while using only a few parameters. In this talk, will briefly summarize my research on physics-based simulation of elastic deformation and collision detection. I will first provide brief overview of my Masters.

work on GPU based real-time Continuous Collision Detection and then discuss current research on the topic of simulating geometrically non-linear elastic deformation at interactive rates using model reduction and GPUs

CARMA: A quantitative formal language for modelling collective adaptive systems

In this talk, I’ll introduce the language CARMA which has been developed as part of the QUANTICOL EU project. This project investigates the quantitative modelling of collective adaptive systems. I’ll give a very brief history of various languages that led to CARMA, highlight keys feature of CARMA itself and talk about three different models we have developed in CARMA relating to ambulance deployment, pedestrian movement and food security.

Streaming Communication Patterns for Multidimensional Domain Decomposition Methods

One of the most significant challenges faced by parallel computing is that of efficient communication between processors. This is becoming more and more important as the industry advances towards exascale computing, as many current algorithms will not scale to the required numbers of processors. One such algorithm is halo exchange – a standard communication pattern for problems that are decomposed by domain. This talk will discuss some of the problems with using halo exchange on exascale machines, and also explore the advantages of a stream-based alternative. It will cover the topic of multidimensional decompositions and the challenges they pose to both streaming and exchange patterns.

Compiling Query DSLs over Complex Objects to Spark

Assume you have a query over complex data. It’s lots of data, so you want to distribute execution. What do you do? Write Spark code. What, by hand?  No. You compile your query to Scala code that calls Spark APIs for you, obviously. Bonus points for a compiler written in Coq. That’s what I did at IBM last summer.

Can quantum theory be characterized in terms of information-theoretic constraints?

Does information play a significant role in the foundations of physics? We investigate whether information-theoretic constraints characterize quantum theory. In a C*-algebraic framework, this is known to hold via three equivalences: no broadcasting and noncommutativity; no bit commitment and nonlocality; no signalling and kinematic independence. But this complex linear framework could be said to build in quantum theory from the start. We show that the first two equivalences break, and the third holds, in a framework of generalized, possibly nonlinear, C*-algebras. This framework of monoidal categories covers arbitrary concurrent computations.

Modelling and analysis of spatio-temporal properties of stochastic systems

In this talk I will present a novel formal framework to model and study spatio-temporal properties of stochastic models of ecological systems. This formal framework consists of a new stochastic process algebra MELA, for Modelling in Ecology with Location Attributes, and suitable techniques to analyse properties expressed through suitable spatio-temporal logics. The aim of MELA is to provide ecologists with a straightforward yet flexible modelling tool, allowing the modeller to create stochastic, individual-based and spatially explicit models. Using stochastic simulations describing the dynamic evolution of the systems, we perform statistical model checking to analyse properties of the models, formally expressed through the recently proposed Signal Spatio-Temporal Logic (SSTL). I will explain how this has been extended to widen the analysis of stochastic systems, introducing a case-study to show interesting examples of applications of this new logic.

Communication, computers, and the chemistry of life

What do chemicals, creatures, and computers all have in common? Communication. For a species that spends most of our time messaging, emailing, tweeting, poking, and just talking, we give relatively little thought to what communication really is. In this talk I will attempt to remedy this by introducing you to process algebra, a mathematical theory which captures the key algebraic features of communication, and abstracts away all of the annoying social and ethical details. We will then see how everything a computer can do can be accomplished by talking alone, how to model competition of species as communication, learn what language chemicals speak, and how to talk ourselves into knots.

Rethinking On-chip Caches for CMP Servers

Multicore servers dedicate close to half of their chip area for a large cache pool. On-chip caches prevent frequent slow off-chip memory accesses making them essential for good system performance. Cache capacity requirements grow in line with the growing application working set sizes. This presents a two-fold problem: on-chip communication delay increases with cache size making cache accesses slow, and secondly the available cache capacity is rendered insufficient. Commercial workloads are sensitive to both cache size(on account of their large working sets) and cache access latency(on account of limited memory level parallelism). There is a need for large caches without compromising on access latency. We propose to employ vertical integration to provide large cache capacities very close to the operating cores for fast accesses. In this talk, I will present a logical arrangement for on-chip stacked DRAM caches which minimizes access latency while still providing large capacities per core.

A hybrid commutativity-based analysis framework for detecting parallelism

Parallelizing compilers have mostly exhausted any benefits of static analyses, largely based on dependence, used to detect parallelism in sequential legacy source code. Using a hybrid static-dynamic framework based on commutativity and liveness information, allows us to overcome traditional limiting factors and provides a way of expressing more structured patterns of parallelism.