1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
|
\documentclass[10pt]{article}
\usepackage{fontspec}
\usepackage{csquotes}
\usepackage{polyglossia}
\setmainlanguage{english}
\usepackage[backend=biber,style=ieee,maxnames=8]{biblatex}
\renewcommand*{\bibfont}{\small}
\usepackage{titling}
%\renewcommand*{\bibfont}{\footnotesize}
\addbibresource{bib/csa.bib}
\title{Suren A. Chilingaryan}
\preauthor{}
\postauthor{}
\author{}
\predate{}
\postdate{}
\date{}
%\posttitle{\par\end{center}}
\setlength{\droptitle}{-4em}
\begin{document}
\maketitle
I am specializing in the domain of high-performance and heterogeneous computing, computer architectures, and parallel algorithms. On a technical side, I have experience in performance analysis and optimization, parallel programming, low-latency communication, and cloud platforms. Working at Institute of Data Processing and Electronics (IPE) at KIT, I apply these technologies to build software instrumentation for distributed data acquisition and control systems.
I built and are currently maintaining a highly-available cloud platform for KATRIN (KArlsruhe TRItium Neutrino) data acquisition and slow-control systems~\cite{katrin2015detector,katrin2018first}. Parts of the system are adapted to support ASEC (Aragats Space Environmental Center) and SEVAN (Space Environmental Viewing and Analysis Network) particle detector networks in Armenia to study thunderstorm phenomena~\cite{csa2009sevan, chili2010thunderstorm}. I led a software work package of the UFO (Ultra Fast tOmography) project aimed to build a novel instrumentation for high-speed synchrotron imaging with online reconstruction and image-based feedback loop~\cite{kopmann2017ufo}. We developed a control system integrating the beam line devices with a GPU-based image-processing cluster and steering the data from the cameras until the storage~\cite{stevanovic2015concert}. IPE is actively designing novel electronics for multiple collaborations~\cite{caselle2013camera,caselle2014kapture}. Our group is looking for a hybrid solutions coupling the high-speed electronics with fast, but flexible software running on GPUs and other parallel accelerators~\cite{vogelgesang2016dgma}. For instance, recently we have performed a case-study aimed to evaluate the possibility of building the next generation of CMS track trigger using GPUs with round-trip latency below 6 us~\cite{mohr2017cms}.
It is a challenging task to design an efficient computing system. Well designed data flow, a hierarchy of intelligent caches, and efficient parallel algorithms can drastically reduce required investments. Throughout all projects, we take a holistic approach to understand project requirements, identify bottlenecks, and optimize performance-critical components. To get an in-depth understanding of available parallel architectures, I have systematically applied micro-benchmarking techniques. It allowed to find multiple undocumented properties of the available hardware and to develop a range of techniques to balance the load between different computational and memory units achieving higher hardware utilization~\cite{csa2018sbac}. We have developed a pipelined image processing framework and contributed parallel algorithms addressing various hardware platforms including IBM Power, Intel Xeon Phi, and multiple GPU architectures~\cite{csa2011pyhst,vogelgesang2012ufo,ashkarin2015,rshkarin2015,cavadini2018upiv}. To enable interactive remote visualization of large tomographic volumes, we develop a web-based visualization framework combining client- and server-side rendering techniques~\cite{ntj2017wave}. Because of the client-side component, high interactivity is achieved with only small investments in the data center hardware. On the other hand, the server-side component allows to improve quality on demand and makes visualization possible also for slow hand-held devices.
\printbibliography
\end{document}
|
