Apr 1, 2000
In January 1999, CERN began to develop CASTOR ("CERN Advanced STORage
It is an evolution of SHIFT, which has been used at CERN for all experiments
in the 1990's. CASTOR is more scalable than SHIFT and should be able to handle
data for the overlapping run of the NA48 and COMPASS experiments, the latter
starting in 2000. The sustained data rate for NA48 is > 25 MB/s while for
COMPASS it will be > 35 MB/s.
Scalability issues are solved by using multi-threading and database technology
instead of a flat file for the stager catalog.
The first tests show that we get really the full bandwidth of the drives and
of the network even using PCs running Linux as low cost disk/tape servers.
The software can be installed on any platform running Unix or Windows/NT.
The CASTOR user interface is fully backward compatible with SHIFT and the
product itself is very modular, so we will be able to put it in production
component by component over the winter 1999/2000.
Note that the modular and layered nature of CASTOR could permit easier sharing
of software components between HEP sites.
The functionality has been also greatly improved by providing a Hierarchical
Storage Manager facility: a name server has been implemented, it is capable of
using different databases (Cdb, Raima or Oracle) and a Tape Volume Allocator
is currently being developed. The first tests of the name server show that
we get excellent response time even for administrative tasks like listing large
A "Mock Data Challenge" test (7 days at 100MB/s sustained, i.e. around 60TB)
will be attempted for ALICE in February 2000 using HPSS and CASTOR and we will
report on the results.
|Presentation: ||Short Paper: |