* Tape1Disk[01] classes
Kors - is there "double counting" for Tape1Disk0 and Tape1Disk1 classes?
Artem - no. It could be a common space or separate space, but no double counting and data replication.
* Transition Tape0 -> Tape1
Kors - problem for SARA.
Artem- It's a problem for many systems. LHCb has a use case - want to
store the data they produces during release validation (or delete it
without storing to tape). But it seems that using this class can be
avoided by LHCb, either by using Tape1Disk1 class right away, or using
Tape0Disk1 and then physically copying data again using Tape1Disk1 class.
Jos - Do experiments foresee deletion from tape?
Artem - normally no, but sometime.
Kors - SARA never deleted experiment's data in 30 years.
Hellen - FNAL only occasionally deletes data: either test data or
invalidated data.
Artem - BaBar also keeps even oldest MC data. Phisicist do need old data
to compare SW performance etc.
Jos - Who needs tobe responsible for deletion from tape - users or
administrators?
Hellen - At FANL filesare deleted from the namespace only, not actually
from tape automatically. This allows admins to recover the data in case a
user made a mistake.
Lionel - At IN2P3 files will be deleted from tape automatically, no
recovery is supported. Users may use different backup service is
possibility for recovery is desired. Files deleted in dCache also get
deleted from HPSS automatically.
Jos - It must be a site choice
* Discussion on tape vs. disk
Jos - No one should think that "custodial data" on tape is more secure
than on disk.
Hellen - Files get lost on tape rearly. Also tape storage vendor can
recover some files on tapes, reducing the loss even further.
Artem - experiments will replicate data on tape, most of them will have
two custodial locations for each data file.
Kors - Tape is still cheaper and easier to deal with for permanent
storage.
Jos - yes, managing disk is more complex, but experiments could still
consider tape vs. disk.
* variouse topics
Artem - do our meetings bring better understanding of SRM, storageclasses,
experiments' reqiurements?
Kors - people do become aware of problems :)
Kors - Ownership during the class transition? Will it change?
Artem - Should not. Transitions will be permitted to the owneronly? To
confirm with deleopers.
Artem - Next meeting? Most likely during the week of Sep 18
Kors - will talk to Roger Jones from Atlas and invite him to present Atlas
storage.
