Title: Disk-only storage solutions for CMS Analysis and Processing at Tier-0 and
Tier-1 sites

Abstract: The operational experience collected during the past years in distributed
data processing and storage by the LHC experiments has confirmed the expected
limitations of hierarchical storage management. It is evident that production
activities requiring tape access need to be carefully planned and executed to be
efficient, and that tape access transparently triggered by "chaotic" user activities
are likely to have a negative impact on the entire system, in particular with the
large increase in the LHC luminosity.
At the same time substantial disk resources using affordable commodity hardware that
are larger than a Tier1's tape archive are increasingly available. For these reasons
the CMS computing system is starting to restrict tape access, strictly isolating
user activity from the tape system and explicitly managing tape recalls at Tier-1 sites
for organised data processing. This means jobs only access data hosted on large
disk-only storage resources and the tape backend is used as a data back-up resource
without any HSM automation.
In this paper we describe the solutions being investigated at CMS Tier-0 and Tier-1
sites: while EOS has been chosen for all user activity at CERN, different solutions
are investigated at Tier-1 sites, based on Lustre, EOS, dCache, StoRM and CASTOR, as
well as adapting data management and data location tools for better disk/tape
accounting, and the integration into job scheduling mechanisms. Finally, we describe
the Stage Manager, a new CMS tool recently developed to manage all tape staging at
Tier-1 sites and some preliminary operational experience.

Authors: Daniele Andreotti, Diego Da Silva Gomes, Gerard Bernabeu, Chris Brew,
Dave Evans, Josep Flix, Sebastien Gadrat, Chiara Genta, Stephen Gowdy, Burt Holzman,
Andrew Lahiff, Syue Yi Liao, Simon Metson, Oliver Oberst, Andreas Oehler,
Emmanuel Olaiya, Andrea Sciaba`, Shu Ting Tiao, Marian Zvada