624
M. Delfino
Information Servers) is a relatively high overhead and latency for file open/close
operations and the loss of full POSIX compliance. In some cases, applications
have to use input–output methods specific to the data server product, such as
Xrootd, 47 whereas in other cases standard protocols, such as NFS or http or
subsets of POSIX I/O can be used. Historically, namespaces have been confined
to single clusters. There are, however, some interesting implementations of global
namespaces, using for example a hierarchy of distributed name servers coupled
through redirection 48 techniques offered by various protocols such as Xrootd,
WebDAV or http. 49 Many commercial systems use similar implementations, but
with different design criteria, for example the requirement of serving hundreds
of thousands of users at relatively low performance in a social network.
Tape servers are hidden from the user, 50 except in very special cases such
as data recording of experimental data. Nevertheless, users and system managers
should be vigilant, as the user may generate access patterns which make extremely
inefficient use of the underlying tape cartridge system. Tape servers and their
related tape cartridge systems are quite complex technologically. Exploring this
is beyond the scope of this chapter, but a few remarks are appropriate. The main
motivation for introducing Tape servers is that they provide storage which is less
expensive than Disk servers, with additional features such as guaranteed readonly access, protection against erroneous deletion by users, lower probability of
data loss due to hardware failure and lower power usage. As their name implies,
tape servers have been historically implemented using storage on magnetic tape
cartridges. Implementations for particle physics are of the “Active Tape” nature,
using specific software to manage the tape in a much more agile manner than
traditional tape backup systems, as well as providing file by file access. Different
computer centers use different software packages for Active Tape management.
Examples are HPSS 51 and Tivoli Storage Manager, 52 both developed by IBM,
Enstore 53 developed by Fermilab and Castor2 54 developed by CERN. Decreasing
prices for commodity disks, however, are driving an emerging area where Tape
47 Dorigo, Alvise, et al. “XROOTD-A Highly scalable architecture for data access.” WSEAS
Transactions on Computers 1.4.3 (2005).
48 For example, redirection is a standard feature in the http protocol used for the World Wide Web.
49 See for example Bloom, Kenneth, and Cms Collaboration. “Cms use of a data federation.”
Journal of Physics: Conference Series. Vol. 513. No. 4. IOP Publishing, 2014.
50 Conventional backup to tape of files in home and project directories is usually performed as part
of the general operation of clusters using standard commercial or open-source packages. This is an
important function but, being completely conventional, is not considered further in this context.
51 Teaff, Danny, Dick Watson, and Bob Coyne. “The architecture of the high performance storage
system (hpss).” (1998).
52 IBM Corporation, Tivoli Storage Manager; see http://www.ibm.com/software/tivoli/products/
storage-mgr/
53 Bakken, Jon, et al. “Enstore Technical Design Document.” Fermilab-JP0026 (1999).
54 Presti, Giuseppe Lo, et al. “CASTOR: A Distributed Storage Resource Facility for High
Performance Data Processing at CERN.” MSST. Vol. 7. 2007.
M. Delfino
Information Servers) is a relatively high overhead and latency for file open/close
operations and the loss of full POSIX compliance. In some cases, applications
have to use input–output methods specific to the data server product, such as
Xrootd, 47 whereas in other cases standard protocols, such as NFS or http or
subsets of POSIX I/O can be used. Historically, namespaces have been confined
to single clusters. There are, however, some interesting implementations of global
namespaces, using for example a hierarchy of distributed name servers coupled
through redirection 48 techniques offered by various protocols such as Xrootd,
WebDAV or http. 49 Many commercial systems use similar implementations, but
with different design criteria, for example the requirement of serving hundreds
of thousands of users at relatively low performance in a social network.
Tape servers are hidden from the user, 50 except in very special cases such
as data recording of experimental data. Nevertheless, users and system managers
should be vigilant, as the user may generate access patterns which make extremely
inefficient use of the underlying tape cartridge system. Tape servers and their
related tape cartridge systems are quite complex technologically. Exploring this
is beyond the scope of this chapter, but a few remarks are appropriate. The main
motivation for introducing Tape servers is that they provide storage which is less
expensive than Disk servers, with additional features such as guaranteed readonly access, protection against erroneous deletion by users, lower probability of
data loss due to hardware failure and lower power usage. As their name implies,
tape servers have been historically implemented using storage on magnetic tape
cartridges. Implementations for particle physics are of the “Active Tape” nature,
using specific software to manage the tape in a much more agile manner than
traditional tape backup systems, as well as providing file by file access. Different
computer centers use different software packages for Active Tape management.
Examples are HPSS 51 and Tivoli Storage Manager, 52 both developed by IBM,
Enstore 53 developed by Fermilab and Castor2 54 developed by CERN. Decreasing
prices for commodity disks, however, are driving an emerging area where Tape
47 Dorigo, Alvise, et al. “XROOTD-A Highly scalable architecture for data access.” WSEAS
Transactions on Computers 1.4.3 (2005).
48 For example, redirection is a standard feature in the http protocol used for the World Wide Web.
49 See for example Bloom, Kenneth, and Cms Collaboration. “Cms use of a data federation.”
Journal of Physics: Conference Series. Vol. 513. No. 4. IOP Publishing, 2014.
50 Conventional backup to tape of files in home and project directories is usually performed as part
of the general operation of clusters using standard commercial or open-source packages. This is an
important function but, being completely conventional, is not considered further in this context.
51 Teaff, Danny, Dick Watson, and Bob Coyne. “The architecture of the high performance storage
system (hpss).” (1998).
52 IBM Corporation, Tivoli Storage Manager; see http://www.ibm.com/software/tivoli/products/
storage-mgr/
53 Bakken, Jon, et al. “Enstore Technical Design Document.” Fermilab-JP0026 (1999).
54 Presti, Giuseppe Lo, et al. “CASTOR: A Distributed Storage Resource Facility for High
Performance Data Processing at CERN.” MSST. Vol. 7. 2007.