Christopher Hollowell
(BNL)
10/11/06, 9:00 AM
The BNL RHIC/ATLAS Computing Facility (RACF) Central
Analysis/Reconstruction Server (CAS/CRS) Farm is a large
scale computing cluster currently consisting of ~2000
multiprocessor
hosts running Scientific Linux. Besides providing for
computation, the CAS/CRS systems' local disk drives are used
by network
distributed data systems such as dCache, ROOTD and XROOTD to
store considerable...
Bogdan Lobodzinski
(DESY)
10/11/06, 9:30 AM
TORQUE is a successor of the OpenPBS batch queuing system,
available as an Open Source product. Despite the wide spread
usage of TORQUE as Job Management System on computational
farms and LHC grid installations, this batch system does not
support any advanced authentication mechanisms.
We show two possibilities, how to redesign the existing
source code in order to add Kerberos 5...
Bryan Hess
(Jefferson Lab)
10/11/06, 10:30 AM
The upgrade to Jefferson Lab will require a hardware refresh
of the mass storage system in order to handle the higher
volume of data from new experiments and simulations. The
next generation, higher capactity tape drives are also
significantly faster, a fact that has implications for
almost all parts of the mass storage system. This talk
examines the performance tuning required to make...
Carl Timmer
(Jefferson Lab)
10/11/06, 11:00 AM
The author describes his recent experience porting software
packages to and running these packages on 64 bit machines
with Solaris and Linux. Issues discussed include code
modification, compiling, operating system requirements, and
performance comparisons with 32 bit machines.
Tom Langley
(NERSC)
10/11/06, 11:30 AM
I would like to explain a bit about our global filesystem
and it's use on PDSF. Also about how this filesystem can be
extended to other sites/labs. Our filesystem is GPFS, but
the concept can also be extended to Lustre or other cluster
filesystems.
Jonathan Schaeffer
(CC-IN2P3)
10/11/06, 1:30 PM
Storage Classes attempt to represent storage use cases for a
given experiment. It is considered harmfull to match the
storage classes to real life storage system especialy if the
latter is based on path to get the storage configuration of
a file.
This presentation aims to define the problematic of Storage
Classes, explain one possible solution which is implemented
at CCIN2P3 and...
Helge Meinhard
(CERN)
10/11/06, 2:00 PM
This talk will present the current state of the art of
benchmarking at
CERN. We will explain our benchmarking procedures, review our
latest results and talk about where we are going from here.
As part of the results review, we will comment on the
current CPU trends and we will talk about the increasingly
important power consumption.
Tony Cass
(CERN)
10/11/06, 2:30 PM
This talk will describe some improvements to the monitoring
and management of the storage and CPU services in the
following areas
- use of SMART for disk monitoring
- integration of disk server monitoring and storage system
management
- transmission of Grid job memory requirements to the local
workload management