# QoS WG Meeting
https://indico.cern.ch/event/850176/
Monday 23rd Sept
# Participants
Oliver
Martin Barisits
Mario Lassnig
Julia
Mihai
# Review of survey conclusions
The only way to gain cost saving is to ditch redundancy and go to jbod
Mario - sites say that operating costs are not the major cost
add this explicitly to the report
WAN issue - caching is supposed to be a remediation, but maybe that's not going to save any money.
atlas transfer 2-3PB WAN
most are done because we see free CPU
so it's reactive
access is 10-12PB a day
CMS privilege WAN access rather than transfer
Is a QoS area for WAN access worthwhile
with extra throttling?
Add comment on SSD pool
categorise the use cases
Maybe a comment about the diversity of systems deployed
posix, ssd, grid
# Actions
if we suggest HEPiX to progress the site level issues, need a channel between them and the WG
what's the involvement of storage providers in all this?
maybe start identifying classes from workflows
Julia endorses the experiment consultation approach
get a standard set of questions up front so the answers can be compared
Move medium term towards an experimention between a site and an experiment.
Grab 2 or 3 sites with different QoS areas and understand how we can use them.
Need a process of recognition/classification of existing QoS diversity and an invitation to experiment
Free for all? sites encouraged to make (small) experimental installations available with a certain label. The new QoS could be "experimental" for instance.