ATLAS/GridPP poster

1
ATLAS and GridPP
GridPP Middleware
GridPP
GridPP is a collaboration of Particle Physicists
and Computing Scientists from the UK and CERN,
who are building a Grid for Particle Physics. UK
Physicists are currently preparing for the Large
Hadron Collider (LHC) that will turn on in 2007
at CERN and produce an enormous stream of data
(millions of Gigabytes per year) that must be
stored and processed (using up to one hundred
thousand processors). No single computer centre
will be able to provide both the storage and
computing facilities for the entire LHC
operation. Hence distribution of computation and
data via the Grid are essential. The UK expects
to play a major role in the analysis of data from
the LHC and the exciting discoveries that are
anticipated. UK Physicists are also
participating in a number of US-based experiments
that are already producing data. Although not yet
on the scale expected from the LHC, these
experiments are using early Grid developments as
a practical tool for doing real analysis today.

• There are three main developments within GridPP
  
• Grid software (middleware)
• Grid-enabled applications and
• provision of computing infrastructure in the UK
  and CERN.
• GridPP will enable testing of a prototype Grid of
  significant scale, providing resources for the
  LHC experiments ALICE, ATLAS, CMS and LHCb, the
  US-based experiments BaBar, CDF and D0, and
  lattice theorists from UKQCD.
• Prototype middleware is being developed in the UK
  as part of the EU DataGrid project and is
  illustrated by
• Dynamic Grid Optimisation where strategies are
  developed to make best use of all the available
  resources across the Grid
• R-GMA used to access information services
  essential to Grid operation. The Relational Grid
  Monitoring Architecture. Producers register with
  the registry and describe the type and structure
  of information they want to make available to the
  Grid. Consumers query the registry then contact
  the producer directly to obtain the relevant
  data.
• GridSite which enables members to update the
  GridPP central web service using Grid certificate
  authentication.

Simulation of GridPP Testbed
Job times for different replication policies
10000 simulated jobs.

• GridPP also funds the Hardware for the prototype
  Tier1A facility based at RAL. This has 400kSI2k
  at present, 80Tb of usable RAID disk and a 180Tb
  tape-based tapestore. In the recent Data
  Challenge 1 for ATLAS, the UK Tier-1 and Tier-2
  facilities represented the second largest
  available CPU resource.

GridPP Applications for ATLAS
European Data Grid Integration
Grid User Interface for ATLAS LHCb
The UK has been integrating and validating the
EDG grid middleware for ATLAS. Initial tests of
EDG release 1.2 on the core sites at CERN, CNAF,
Lyon, NIKHEF, RAL by the ATLAS-EDG group revealed
many problems with Resource Broker saturation
and the use of a single Replica Catalogue (solved
with RLS). The job success rate was only 70.
GANGA/GRAPPA is a project working to produce an
interface between the user, the Grid Middleware
and the experimental software framework.It takes
advantage of the shared software framework in the
two experiments. It is being developed jointly
with the LHCb experiment, and as it is using
component technologies will allow reuse elsewhere
It was started within GridPP but is a vital
partnership with US ATLAS.
Site RAL Cambridge I C Birmingham
Jobs Allocated 60 180 40 20
Success 57 170 40 17
A more recent UK mini production used input data
stored on RALs tape server, the requirements in
JDL and the IC Resource Broker and boxed set
executables. The test took only 1 week, 1
operator and 3000 SpecInt95 days. The success
rate was higher than 90 . While not yet suitable
for production, this is an encouraging step
towards a brokered production system.
The GANGA internal architecture
GridPP also contributed to AtCom, tool used for
DC1 production, and which acts as a testbed for
GANGA developments. The UK are responsible for
the plug-ins for each batch and Grid system.
Packaging, Installation, Configuration An
important issue is user software installation.
The large number of sites, many serving diverse
user groups, requires automated and scalable
Installation Tools. We create coherent rpms and
tar files from CMT (which maintains the software
and the runtime environment), which exposes the
package dependencies in the form of cache files.
These are then used by PACMAN which can either
pull or push complete installations to remote
sites. Scripts are available to make the process
semi-automatic.

• User input has proven essential to building the
  system.
• Jobs must make as few assumptions as possible
  about the system.
• Configuration and integration takes as long as
  writing the middleware.
• Inter-operability between Grids will be a
  challenge for us all.