Advanced Computing Research Centre
|
|
|
| ACRC home | ||
| About HPC | ||
 BlueCrystal Phase 3 |
||
| BlueCrystal Phase 4 |
||
| Application Form |
||
| Performance |
||
| Policy |
||
| Members | ||
| Training and Support | ||
| Storage | ||
| News and Events | ||
| People |
Performance
Examples of Improvement in Job Run Times using BlueCrystal
| Name and Department | Research Area | Improvement in run time/performance |
| Dr Marco Turchi Department of Engineering Maths |
This project is part of the European project Smart: Statistical Multilingual Analysis for Retrieval Translation, the goal being to evaluate the quality of Statistical Machine Translation (SMT) on different datasets and parameters. Training and optimization of the SMT system are very expensive in terms of time, in particular when using huge amounts of training data, i.e. 22 million sentences. | On a desktop the maximum number of sentences that could be run is 1.1 million, whereas on BlueCrystal due to the large number of processors available and the parallelisation techniques employed on the code, 22 million can be run. |
| Dr Tom Gaunt and Christopher Raistrick School of Social and Community Medicine |
The STEPs project investigates the effect of a category of human genome variants on the modification of DNA sequences which can lead to changes in gene expression and influence disease risk. The cost of calculating each pairwise correlation between variants is high, with ~2340 files each containing ~1million pairs of variants. | This cut down over a decade of work on one quad-core machine to a few months coding and executing on BlueCrystal. |
| Dr Beate Glaser Department of Social Medicine |
The study assessed the empirical evidence between genetic variation within loci of the G-protein family and intelligence using genome-wide SNP data. For this 10000 simulations were set up and run using Plink software. | Reduced the computing time from around 14 days on a stand alone machine to less than 6 hours on BlueCrystal. |
| Dr Silvia Perez Espona Department of Biological Sciences |
The project is assessing the population structure of the neotropical army ant Eciton burchellii and its co-evolution with some of their guests. Ran the program STRUCTURE to assess the population structure of this army ant in Panama and Costa Rica. | Reduced the computing time from 2 months on a stand alone machine to less than 1 day on BlueCrystal. |
| Dave Platt Department of Mathematics |
Developing new algorithms for verifying the Riemann Hypothesis (RH), with the immediate goal of testing RH for all Dirichlet character L-functions of conductor up to 100000 and height up to 1000. Calculated about 3 billion values of the Hurwitz Zeta function to high precision (300 bits). | On a twin core desktop, it would have taken about 17 days, but using 4 nodes, each of
8 cores, this calculation was done in 1 day. If extrapolated to the whole computation, initial indications are that BlueCrystal might reduce an 11 year calculation down to about 3 months. |
| Dr Christopher Woods School of Chemistry |
Developing a new methodology for use in computational drug design. Rapid turn-around of simulations is needed to see if a change to the method is an improvement. |
If the simulations were run on a desktop they would have each taken about 20-40 days. On BlueCrystal they all took about
1 day. Because several simulations could be run at once, about 20 different simulations were run in the set over a period of 7 days, which would have taken about 1.5 years to run on a desktop. |
Professor Frank Windmeijer Department of Economics |
Simulated the distribution of the dissimilarity index, a measure of segregation, indicating for example how segregated by poverty the schools are in an area. Used the multi-core multi-threading version of Gauss, and ran multiple instances at the same time. | Reduced the computing time from 4 weeks on a stand alone machine to 1 day on BlueCrystal. |