Overview
========

Questions about DPM gathered by Frederique Chollet.

Answers by:
	- Jean-Philippe Baud (jean-philippe.baud@cern.ch)
	- Sophie Lemaitre (sophie.lemaitre@cern.ch)


Questions / Answers
===================

- DPM architecture - SRM request handling
	- See tutorial


- Namespace configuration, database description, synchronizition with disk servers
	- See tutorial
	- Consistency scripts - First version will:
		- List missing physical files, present in DPM Name Server
		- Give possibility to remove files from Name Server


- Data management and security : restoring data, data migration, coherency cheking at all levels
	- restoring physical files: need disk backup (incremental backup or RAID)
	- restoring logical files: need database backup (for cns_db and dpm_db)
	- data migration: dpm-drain utility


- Pools management : evolutions foreseen (draining tool, how to handle VO quotas)
	- VO quotas are on the TODO list (might be there before the end of the year)


- VOMS support
	- VOMS is supported


- VO dedicated pool
	- proposal for pool support of a group list.
	- solution designed, not implemented


- Current situation and problems with ALC. ACL handle by the VO?
	- We are in a transition phase (from no VOMS support to VOMS support)
	- Thus, kind of migration period
	- Need to update ACLs on *existing* directories and files (because of transition)
		- if lots of existing entries, DPM admins should do it using the (fast) tool we provide
		- if very few entries, can be done by users with "lfc-setacl"
			=> RESTRICTION: to use lfc-setacl, the user has to be the *owner* of the file/directory
			=> No concept of VO admin
	- For new directories, Experiments should set the ACLs themselves
		- With lfc-setacl
			=> RESTRICTION: to use lfc-setacl, the user has to be the *owner* of the file/directory
			=> No concept of VO admin
	- Secondary groups support will improve the situation
		=> IMPORTANT: it will not solve everything, ACLs still have to be set


- DPM scale and optimisation

- Is it recommended to aggregate disks from WNs ?
	- We have no experience with that
	- All WNs have to be trusted by DPM => no WNs where users can become root !
	- Quality of Service ? WNs stability usually not that good (easily rebooted, scratched, etc.)
	- Volatile files on WNs -> might be OK
	- Permanent files on WNs -> NOT OK


- Limit in pool number?
	- 59 pools maximum


- Limit in FS per pool per DPM?
	- No limitation from DPM software
	- But limit in memory size...


- Limit in GB for the DB?
	- No limitation from DPM software
	- Depends on DB configuration


- Recommandation for better performance?
	- DPM server should not be on a disk server.
	- for big systems, does MySQL scale ? We are not sure... But Oracle is also supported.
	- we are currently understanding performance problems with many simultaneous transfers, or many simultaneous opens.


- GRIF federation long term project of unified SE with distributed disk servers
	- Disk servers can already be on different domains
	- No plan to support distributed file systems
	- Someone from CEA sent something about this => we would need more documentation


- Implications of SRM v2 implementation and corresponding Storage Classes :
How space-token will work? Migration plans ? Evolution of the current information published ?
	- SRMv2.2 released in DPM 1.6.3
	- Storage Classes -> space tokens defined by admins (or users)
	- No migration:
		- Existing files will not be moved to spaces
		- "srmChangeSpaceForFile" SRMv2.2 method defined, to be implemented this year (WLCG decision)
	- Static space reservation -> ADMIN with "dpm-reservespace" (no limit)
	- Dynamic space reservation -> USER with "dpm-reservespace" or "srmReserveSpace" (limit on duration and size of reservation)
	- Information provider now publishes SRMv2.2 endpoint
	- Admins can publish space tokens in Info Sys but:
		- No client software would use it at the moment (neither WLCG, nor Experiments software)
	- Note: space tokens are use only for "Put" operations, and ignored for "Get" operations (WLCG decision)


- Current DPM bugs open
	- See Savannah and GGUS


- TODO list
	Priorities are (not ordered):
	- Finish and release secondary groups
	- Implement pools reserved to a list of groups
	- Release with GridFTP2 (for SLC4)
	- DICOM interface for BIOMED (extension of the DPM)
	- SLC4 on 64 bits
	- Consistency scripts