Cost estimation for sites

→ Europe/Zurich
513/1-024 (CERN)

513/1-024

CERN

50
Show room on map
Registration
Participants

Minutes from Renaud

--------------------------------------

We need milestones.
A good start would be to have a simple estimate from sites first, then go to further detail.
We need to include manpower at some point, but we should concentrate on computing resources first, because future computing models are being built based on that.
Then see how to include human resources.

Bernd has shown estimates including the building as well (is that public??)
Ideally we need to include everything, from hardware to power, manpower and building.

UK did an exercise to calculate data center costs, which led to have changes in the number of sites  : there may not have done any model for that, but some knowledge can be taken from that experience.

IOps has a large impact on the cost estimate, which would lead to a better comparison between storage solutions.
But need to take into account whether a tape system is used as hot or cold storage : in the latter case, IOps is almost irrelevant -> we need that input from experiments.
We should take iops into account at some point, even if we don't include it for first iterations.
Feedback from tape performance working group will be valuable.

Feedback from us to is important for the experiments.

Manpower:
we could separate FTE that can go and FTE that can't go.

Hot topic for first milestone :
Need to discuss with sites how to share the numbers in a confidential way, and see how they react with giving numbers.
Need to contact T1s to get the numbers, or to know why they can't give them.
Renaud will takes care of that.
We need first to agree on how to present the questions to the sites :
detailed questionnaire is what we need (understand metrics and measurement method) in the end.
Another approach would be to start with a simple question like 'what would be the financial consequence to change some of your disk storage into tape storage, or vice versa.'
The problem with a statement at a given time doesn't make it possible to do predictions ; for that we need to see how cost evolves for a given resource.

At some point if we have sites blocked by NDA, like CERN, we have to live with it -> include in the model only sites that share data, otherwise we can't do much.

Someone from the WG may get the data from sites and commit not to disclose results outside of the WG.
But for the moment we don't know how sites will react, wait and see the replies.

Licenses:
It would be nice to have an idea on how proportional licenses are dependent on volume.

Networking:
probably hard to get precise results. but summing networking expenses and normalise to some capacity should be possible.
IN2P3 is working on global numbers normalised to get euro/{TB,HS06}/gpbs interconnect
Need to keep WAN and LAN expenses separate.

CERN tape financial breakdown is very detailed, but nothing can be shared -> not sure we can do much with it.

Helix Nebula is using ECAR to compute TCO.
Our WG is still far from that step, we may come back to ECAR at some point.

We will create an egroup for this subworking group and try to organize regular meetings, most likely after the summer break.

 

There are minutes attached to this event. Show them.
    • 09:30 → 09:50
      Introduction and overview 20m
      Speaker: Renaud Vernet (Centre National de la Recherche Scientifique (FR))
    • 09:50 → 10:10
      Resource cost calculation : where we are now 20m
    • 10:10 → 12:00
      Discussion 1h 50m
    • 12:00 → 12:30
      Organization 30m