Speaker
G. Lo Re
(INFN & CNAF Bologna)
Description
Next generation high energy physics experiments planned at the CERN
Large Hadron Collider is so demanding in terms of both computing
power and mass storage that data and CPU's can not be concentrated in
a single site and will be distributed on a computational Grid
according to a "multi-tier".
LHC experiments are made of several thousands of people from a few
hundreds of institutes spread out all over the world. These people,
according to their collaborations on specific physics analysis
topics, can constitute highly dynamic Virtual Organizations rapidly
changing as a function of both time and topology. The impact
of future experiments on Wide Area Networks (WAN) will be non
negligible especially for what concerns the capillarity of bandwidths
(down to the "last mile"), quality of service, adaptivity and
configurability.
In this paper we report on a series of multi-site data transfer tests
performed within the ALICE Experiment on a wide area network test-bed
in order to spot possible bottlenecks and pin down critical elements
and parameters of actual research networks.
In order to make the tests as realistic as possible, reflecting the
real use cases foreseen in the next future, we have taken into
account all the aspects of the elements involved in the transfer of
a file:
- Local disk Input/Output (I/O) performance;
- I/O block size;
- TCP parameters and number of parallel streams;
- Bandwidth Delay Product (BDP) expressed as the product of the
Bandwidth (BW)
times the Round Trip
Time (RTT).
Primary authors
Mr
A. Fritz
(University of Houston)
Mr
D. DiBari
(Bari University)
Mr
D. Mura
(INFN Cagliari)
Mr
E. Fragiacomo
(INFN Trieste)
Mr
F. Carminati
(CERN)
Mr
F. Minafra
(INFN Bari)
Mr
G. Donvito
(INFN Bari)
G. Lo Re
(INFN & CNAF Bologna)
Mr
J. Svec
(ASCR Prague)
Mr
M. Masera
(Torino University)
Mr
M. Sitta
(INFN Torino)
Mr
P. Cerello
(INFN Torino)
Mr
R. Barbera
(Catania University and INFN)
Mr
R. Turrisi
(INFN Padova)
Mr
S. Bagnasco
(INFN Torino)