Speaker
Wilko Kroeger
(SLAC)
Description
The BaBar Experiment stores it reconstructed event data in root files
which amount to more then one petabyte and more then two million files.
All the data are stored in the mass storage system (HPSS) at SLAC and part
of the data is exported to Tier-A sites.
Fast and reliable access to the data is provided by Xrootd at all sites.
It integrates with a mass storage system and files that are not on disk
are automatically staged to disk if requested by a user. The built in
fault tolerance allows clients to survive failures, for example a crashed
data server, and eases the maintenance of the cluster. With Xrootd it is
easy to repair a broken machine without any impact on users.
At SLAC we have two Xrootd clusters. One read only cluster with 59 data
servers which is used by analysis jobs. The second is a production cluster
with 16 data servers which is used for reading and writing files. More
then 4000 cpus access these two clusters.
We will discuss the setup, usage and experience we have with maintaining
Xrootd data clusters for a running experiment. We also will describe
current development of connecting Xrootd clusters between Tier-A sites.
| Submitted on behalf of Collaboration (ex, BaBar, ATLAS) | BaBar |
|---|
Primary author
Wilko Kroeger
(SLAC)
