 |
|
DAILY NEWS AND INFORMATION
FOR THE GLOBAL GRID COMMUNITY / AUGUST 4, 2003; VOL. 2 NO. 31
|
Applications:
GRID COMPUTING CENTER KARLSRUHE
INSTALLS 70 TB DISK SPACE
by Uwe Harms
The Computer Center at Research Center Karlsruhe is the German Tier 1
Center
for the CERN LHC (Large Hadron Collider) data. There will be up to 10 such
centers around the world. In July GridKa (Grid Karlsruhe) installed 70
TeraBytes disk space, which will be extended in different steps to 340
TeraBytes. A consortium consisting of Systematics Technology Solutions and IBM
won the European-wide call for tender. Although GridKa is preparing its
infrastructure for the big request in 2007/2008, several other German Research
groups which are analyzing running non-LHC experiments asked for computing
support. Thus it is in production with its grid approach as a computer center
for high-energy physics.
The CERN Project LHC
At CERN they build the LHC accelerator ring with a diameter of 9 km, 100
meters below the earth. Protons will be accelerated and collide with 7
TeV/particles. The start of the four different experiments is scheduled for
2007. Estimates sum up to 8 Petabytes/year of measured experimental data.
About 8000 elementary particle physicists worldwide will analyze these huge
volumes of data.
To proceed these data, the scientists developed a multi-tier computing
model:
- Tier 0 CERN, gaining the data
- Tier 1 The global LHC Computing Center consisting of 10 centers worldwide
including CERN
- Tier 2 100 University and Lab computing centers
- Tier 3 1000 Institute computers
- Tier 4 the end user at his desktop
Some of the Tier 1 centers are for example University Tokyo, Japan,
Rutherford
Lab, United Kingdom, IN2P3, France, CNAF, Italy, Fermi Lab, BNL, USA, and
GridKa, Germany.
The user should access the 10 Tier 1 centers as a virtual computing center
without knowing where the specific data for his analysis and the application
programs are stored and which center he uses. This requires a strong effort to
modify all the existing open source grid middleware.
GridKa in Karlsruhe
The German Ministry of Science and Research looked for a Tier 1 Center in
Germany and decided to install it at the Research Center Karlsruhe. Thus
GridKa is preparing its infrastructure and all the middleware for 2008. Today
it offers 230 dual Intel Pentium 3/Pentium 4 processors, 460 processors in
total. They sum up to a peak performance of 900 GFlop/s. In late 2003 the
number of PCs will be doubled. About 50 TeraBytes are still installed, which
have been extended in July by another 70 TeraBytes. For 2008 Klaus-Peter
Mickel, Head of the recently founded Institute for Scientific Computing,
expects 22.000 to 23.000 PCs, 1,4 PetaBytes disk and 3,5 PetaBytes tape
space.
70 TeraBytes disk space growing to 340 TeraBytes
To start with the storage issue, GridKa prepared a European-wide official
call
for tender. Starting with 70 TeraBytes the disk space should be extended in
different time steps up to 340 TeraBytes. The bid won Systematics Technology
Solutions, an system integrator, and IBM which delivers the hardware of the
disk storage. Both enterprises are well known within the Research Center
Karlsruhe. In the past IBM delivered mainframes and recently an IBM Power4
machine. Systematics Technology Solutions was a software partner for FZK for
several years. They installed the SAN (Storage Area Network) within three
days. The solution is based on the IBM TotalStorage FAStT700 Storage Server, a
flexible Fibre Channel based solution. The innovative Fibre Channel technology
offers a data throughput of up to 2 Gigabit/s. The system scales very well
with up to 15 FAStT EXP700, expansion units. In it one can find 14 2 Gigabit/s
Fibre channel hard disks. The integrated RAID controller allows the different
RAID levels, RAID 0, 1, 3, 5 and 10. The IBM FAStT Storage Manager supports
the functionality FlashCopy, fast copy of all the stored files without
hindering the usual SAN operations, Dynamic Volume Expansion, on-line
modifying the size of logical volumes without interrupting the operation of
other logical volumes, and Remote Mirroring.
The compute nodes and the SAN are connected via an FC/9000 Fibre Channel
from
Inrange, delivered by IBM. This family of enterprise class Fibre Channel
storage networking directors and switches brings speed, reliability and high
availability.
GridKa
The CERN LHC activities will go into production, but in the meantime the
center is still supporting user groups from all over German. As the German
high-energy Physics community still miss a computer center, they found a
solution in GridKa. In May 2001 the German Particle and Nuclear Physics
community laid down the requirements for a Regional data and Computing Center
in Germany (RDCCG). It contained the requirements for the LHC experiments, but
the RDCCG should also serve other major data intensive non-LHC experiments of
both communities and should extend to other sciences later on. Today GridKa
serves 19 institutions from all over Germany, 41 user groups and about 350
scientists. The center is the production environment for non-LHC experiments
like BaBar at SLAC, Stanford, the experiments CDF and D0 at Fermi Lab, and
Compass at CERN.
The grid functionality is still growing. Up to now there have been only
partly
such functionality e.g. Alice, the LHC experiment. From February 2003 there is
a step-wise building of grid functionality by adding components from all Tier
1 centers. Since 1. July 2003 the Tier 1 centers started a test grid with all
the involved centers.
At Research Center Karlsruhe one of the Research and Development Programs
(F&E) is Grid/Scientific Computing. Four Institutes of FZK are involved in it.
The plans are to analyze meteorological Envisat data in a multi-level
computing grid with distributed capacity. Additional the center will develop
different grid-based applications from medicine, environment and particle
physics.
by Uwe Harms, Harms-Supercomputing-Consulting
|