CoreGRID: A network of excellence in Grid and P2P Technologies

1
The WestFocus Grid Alliance and its Applications
Application of Grid Computing for Designing a
Class of Optimal Periodic Nonuniform Sampling
Sequences
2
The WestFocus Grid Alliance

• Founders University of Westminster and Brunel
  University
• applying the benefits of Grid computing to
  real-world industry applications
• Offering
• Consultancy
• Grid-enabling applications
• Grid technology
• GEMLCA P-GRADE portal
• SZTAKI Desktop Grid technology

3
The WestFocus Grid

• Utility Grid based on state-of-the-art Grid
  technology
• NGS GT2 WestFocus GT4
• Connects three 32 node computer clusters at
  Westminster and Brunel Universities (Kingston
  University will join soon)
• Runs the P-GRADE Grid portal and GEMLCA
• Connected to and interoperable with the UK NGS

4
Application example

• Designing a Class of Optimal Periodic Nonuniform
  Sampling Sequences
• A collaboration between the Department of
  Electronic Systems and the Centre for Parallel
  Computing at University of Westminster

5
Uniform sampling problems with aliasing160MHz
and 40MHz Sinusoids Sampled at 200MSps (Every 5ns)
Large sampling frequency could be a problem in
case of gigahertz signals
6
NonUniform sampling 160MHz and 40MHz Sinusoids
Sampled not more often than Every 5ns
Better results even with lower sampling rates
7
Uniform / Periodic Nonuniform Sampling
Uniform Sampling
Periodic Nonuniform Sampling
T
T
T
T
T
8
Periodic Nonuniform Sampling
Time quantum L Minimum distance between
consecutive sampling instants H, e.g.
H3L Period of the sampling sequence T, e.g.
T17L Number of samples per period N, e.g.
N5 Sampling instants within the first period
t1,,tN e.g. t10, t23L, t37L, t410L,
t513L Distances between consecutive sampling
instants e.g. g13L, g24L, g33L, g44L, g54L
9
How to Measure Quality of PNS?
Aim Minimise J, where
and coefficient
Performance measure J 1 Aliasing is not
suppressed at all J 0 Aliasing is fully
suppressed
10
Optimal PNS Sequence

• Aim
• Find the sequence g1,g2,..gN which minimizes J
• Problem
• When T100L and H5L then the total number of
  possible J values is

3,288,195,610!!!
11
How to Find the Best gi Sequences?

• Let qr, qr1, , q2r-1 be the numbers of gaps of
  sizes rL, (r1)L, (2r-1)L, then
• For each q value find all possible J values
• Number of J values
• When T100L and H5L then the total number of
  possible J values is

3,288,195,610!!!
12
Computational Workload

• Optimization methods search through a carefully
  selected subspace of feasible solutions.
• The subspace increases very fast with larger T
  and H/L.

13
Parallelization
14
Two basic Grid directions
Grid Technology
Utility Grid
Desktop Grid
(UK NGS, EGEE Grid)
15
Grid solution 1 GEMLCA and the P-GRADE portal
P-GRADE Portal Server
16
Creating GEMLCA Legacy code from the DSP
executable
17
DSP Workflow in the P-GRADE portal

18
Grid solution 2 based on SZTAKI DG Technology

• Based on BOINC
• Application development utilised DC API
• More than 100 laboratory PCs utilised currently
• Scales up to thousands of computers

19
Some comparison of the solutions

• GEMLCA/P-GRADE
• User friendly, easy to change parameters
• Scalability problems parameter study support
  is currently added to P-GRADE portal

• SZTAKI DGrid
• Project administration requires BOINC expertise
• Scales up to very large number of PCs

Performance (without comparison)
20
Some comparison of the solutions
21
Thank you for your attention
Contact kisst_at_wmin.ac.uk gemlca-discuss_at_cpc.wmin.
ac.uk Website http//www.gridalliance.co.uk/