Navigation Chart

Main Navigation

Page Content

Theses

Below you find a selectopon of available thesis topics. You can also provide your own topic.

Performance Benchmark Tests 

_______________________________________________________________

The group of Prof. Lindenstruth offers access to several high performance clusters. With simple performance benchmarks the inter process communication shall be benchmarked in terms of performance ans possibilities.

 

 

Auto-tuning of Parameters for Linpack benchmark:

______________________________________________
Linpack benchmark is the standard tool for evaluating the performance of supercomputers.
We have developed an optimized implementation, called HPL-GPU, based on HPL (High Performance Linpack).
This version can make use of GPUs via OpenCL and CUDA APIs.
There are plenty of parameters available which must be tuned manually.
This makes it very difficult to deploy the benchmark at a 3rd party site without knowledge of the benchmarks interiors.
Topic of the work would be to create an autoconfiguration mechanism, which tunes the application for a given hardware.




Combined Track Reconstruction and Track Merging on GPUs for the ALICE High Level Trigger at CERN LHC:

_____________________________________________________________________________________________
We have two individual programs for track reconstruction and track merging on GPUs.
Both programs copy data to GPU and back. It would be beneficial to keep the data completely on the GPU speeding up processing a bit.
Topic of the work would be to create a new HLT component which merges the two existing ones, and changes the software such that data can remain on GPU.




Loop finder for ALICE HLT Track Reconstruction:

__________________________________________
We have a program for track reconstruction on GPUs for ALICE HLT.
By construction, our algorithm cannot find particles with low momentum.
There is an algorithm for low-momentum tracking available from the ALICE offline group.
Topic of the work would be to investigate this algorithm, measure its performance, verify that it is fast enough to run in the HLT, and integrate it in the HLT tracking.




Applications of fast failure-erasure-coding:

________________________________________
We have a fast implementation of a failure erasure coding algorithm.
Currently, it is only a library with a low-level interface.
Topic of the work would be to investigate where this library could be applicable and perhaps create some reference implementations.
In fact, this topic is very free. It is about making use of that library in a more practical way.
The student can think of his own ways if he likes.
One first idea would be a raid-module for linux, or an integration in the available raid linux kernel modules.
For the latter case, advanced linux knowledge is required.

 

 

Arbeit am CBM First-Level Event Selector Input Interface.

________________________________________________________________

This work includes the development for the inputinterface of the First-Level Event Selectors (FLES) of future CBM experiments at FAIR. The FLES is a high performance system used for event selection of the whole experiment. Major element of the input interface are FPGA-PCIe-cards that for example realize a connection between compute nodes and the detectors of the experiment. In context of these card several thesis topics are possible. For example the commissioning and evaluation of the cards own DDR3 memory in connection with the PCIe interface.


For all of the above topics knowlegde in C++ is necessary.

If you are interested in of of the above topics or would like to propose an own topic please contact Helvi Hartmann (hhartman [at] fias.uni-frankfurt.de)

To Navigation Chart
empty

Bottom of Page