|
-Curriculum Vitae -Thesis -Publications -Projects -CoPs -Zero Copy Middleware -Xibalba -Patagonia -ECT -Lectures -Student Projects -Personal Homepage -Research Group -Institute -Departement -ETH |
Dr. sc. techn. 2002, Dipl. Informatik-Ingenieur ETH 1996
|
I am now working for: Google Inc.
|
My former address: Computer Systems Institute
|
I am a postdoctoral researcher and lecturer at the Swiss Federal Institute of Technology (ETH) at the Institute for Computer Systems. My doctoral thesis under the supervision of Prof. Thomas M. Stricker in the Parallel and Distributed Systems Research Group focuses the area of parallel and distributed computing with component object middleware (e.g. CORBA) and clusters of commodity PCs (CoPs). I am especially interested in high performance communication systems and high speed networking in clusters where the aim of our work is to improve the overall efficiency in parallel and distributed systems by solving communication performance problems of system software and middleware. Our zero-copy TCP/IP implementation for Linux shows, that an efficient zero-copy regime is difficult to introduce but possible. An extention of the zero-copy paradigm to distributed object middleware (DOM) like CORBA leads to very efficient distributed applications.
After finishing high school in St.Gallen, I worked as an analyst and programmer at Bühler AG Uzwil, after which I started my computer science studies at the Swiss Federal Institute of Technology Zürich. After an internship at Control Design and Development Ltd in Peterborough, England, and a term project about Elastic Deformable Contours in Image Segmentation (available in German) I graduated in 1996 with my Diploma thesis on Compression Domain Volume Rendering for Distributed Environments (available in German) in the area of Fast Wavelet Based Volume Rendering where I implemented parts of the Experimental Volume Visualization Environment - EVOLVE.
A detailed Curriculum Vitae is available over the Web.
The CoPs-Project:
Clusters of PCs, our umbrella project
| Our CoPs-Project investigates architectural and operating system support for parallel and distributed computing on PC clusters interconnected with a Gigabit/sec network. Recent projects unter the umbrella of CoPs are the following: | ||
| |
Zero Copy Enabled Gigabit Ethernet Clusters of Personal Computers (CoPs) offer the best compute performance at the lowest price. Workstations with 'Gigabit networking to the Desktop' can enable a new game of multimedia applications that benefit from higher communication bandwidth and lower latency. In order to reach the full Gigabit/s speed on normal PCs with their typically weak memory subsystems requires either additional hardware for protocol processing or alternatively a highly efficient software system that circumvents data copies. We successfully introduced speculation techniques into system software design and managed to implement a clean zero-copy solution entirely in software that runs with commodity network interface cards (NICs). | |
| | Distributed
Object Packages for CoPs and their Performance (Zero Copy
CORBA). Since the zero copy principle is applicable to all levels of software, we extend the zero copy design from low level drivers to middleware packages like CORBA or MPI. We implemented a prototype of a data intensive application, that is properly modeled by parallel objects in CORBA and still strictly adheres to the zero copy regime of a highly efficient software implementation running on Clusters of PCs. | |
| | Xibalba, our recent work in
computer architecture Xibalba is a 128 x 1GHz multiboot cluster, which is used for a variety of research within the departement of computer science. The systems is based on commodity PC hardware connected by a Gigabit networking technology with good scalability for several applications in scientific computing, distributed databases and everyday computing. | |
| | ECT memperf - Extended Copy
Transfer Characterization ECT memperf is a Memory Performance Characterization Toolkit and Benchmark. The method allows to characterize the performance of memory systems by capturing two aspects of the memory hierarchy. First its behavior with temporal locality by varying the working set size and second the spatial locality by varying the access pattern. | |
| | Patagonia CloneSys
and Dolly CloneSys is a free tool to install multi-boot environments by partition cloning. Using the streaming tool dolly, which implements a multi-drop chain, the tool provides a very efficient method to distribute OS-images. The CloneSys tool is especially usefull for Cluster Installations. It is used at our CS-Departement to maintain more than 100 Multiboot Student-PCs as well as the Xibalba- and CoPs-Cluster. | |
| | A
Comparative Study of Gigabit Networking Technologies for PCI Bus Based
PCs Several commercial Gigabit interconnect technologies are emerging due to the results of the intensive research with the NSF Gigabit Networking testbeds. A fair comparison of these different network technologies needs a common denominator as a basis. We propose to carry out a comparison at three different levels: (1) simple remote load/store operations, (2) message passing libraries and (3) standard IP over LAN networking. Our comparison includes the two Gigabit Interconnects for the PCI bus that are targeted at clusters of PCs, Myrinet and Dophin SCI. Just for a reference the interconnect technology of the SGI/Cray T3D is evaluated with the same benchmarks. | |
Our Computerlab
with View over
Zürich
A view HOWTOs
for the ETH-internal usage of the CS-ZZ-Cluster
Computer Systems Performance Analysis and Benchmarking: Course Homepage
| My course about
Performance Evaluation discusses techniques for experimental design,
measurement, simulation, and modeling. It is based on a text book of
Ray Jain. We will first discuss the importance of the different metrics for the performance of the processor, the memory system, the communication system and the I/O system. After a brief look at a few theoretical tools, we will examine common industrial benchmark suites like the SPEC benchmarks, the SPLASH suite for shared memory parallel systems and the TP benchmark series for transaction processing systems. After completing this course you will have a better understanding of the important performance metrics that characterize modern computer systems, how to tune software for these systems and a sound technical background for equipment selection. | ||
