How will data be managed at T1/T2?



Greig Cowan: Storage group



Issues: 



Data will be lost (disks will fail, will be problems with dcache)

Experiments need to understand this.

Must have fail over capabilities => replication between sites

For example Dcache - How do users know that they can't get problem data

anymore - this is a question to VO's



Quota's - We need a quoting system. Tier 2's currently find this a problem.

When will they come into practise?



(Roger Jones) Quotas need to be assigned at role/VO level







Phil Clark:



T2 disk minimally used. The amount of disk per experiment varies. From a

storage management point of view (maintenance) is it cost effective to keep

disks going until they are really needed?. (Roger Jones - We cannot delay 

until they are needed)



Should we be recommending different spec's (Dcache and Dcache resilient mode)

(John Gordon - this is down to the deployment team)



Jens Jensen: Leading storage group



Interacts with user group, vo's directly and the DTeam.

Works on storage model changes - in general this is working well

He wants to ensure that storage is covered at all sites (including dark storage)

Need to support at least one distributed file system

- fulfill different needs. DPM is more straightforward than the more complex DCache.

With respect to storage Qos - they monitor uptime and it is possible to get passive

logs out of the transfer systems. Also need to analyse throughput/ bandwidth and that

there is high enough robustness

SRM 2.0 is next on roadmap - and how to provide services to T1's and T2's



David Martin (Provided 2 slides on disk storage at Glasgow)

Obtained 10 boxes from clustervision, set up differs from Rutherford

Attempt at doing some burn in:

http://weather.ou.edu/~apw/Projects/stress

http://home.comcast.net/SCSIguy/SCSI_FAQ/RMiller_Tools/dt.html



Gianfranco Sciacca:



Is data loss VO or site Responsibility?

(Roger Jones) VO - T2 disk storage in the long term is not supposed to be backed up

or expected to be completely resilient but site has responsibility.

The ATLAS strategy is different to the CMS strategy

(Graeme Stewart - Sites must be responsible for a reasonable quality of storage)



Brian Davies:



Qos depends on man power to solve problem. There is an issue where one must keep

track of data on the SRM as it evolves over time.

FTS/SRM – When single file system goes – takes time to find details in particular pool.



Paul Millar:



Works with Greig and Graeme on the management side. Specifically with monitoring.

Data management is more than just storage - it is movement of files i.e FTS

FTS on a Tier 2 site - is it needed?

(Tony Doyle - See if T1 model works first - then can make a decision on this)



Conclusion: Need to improve communication between all sides.