Atlas CAP Closeout

1
Atlas CAP Closeout

• Thanks to all the presenters for excellent and
  frank presentations

2
Project Overall

• U.S. Atlas (and International Atlas) have made
  considerable progress in the past 12 months
• Budgets remain very tight, and may jeopardize
  project goals, especially in the out years

3
Management (1)

• Overall, the management of the Software
  Computing of the research program appears
  healthy...
• For FY2006-FY2007, an adequate management reserve
  is being maintained. For FY2008, it is
  ephemeral, in that the computing project deficits
  consume the full reserve (i.e. there is no
  reserve). Moreover, the Tier 1 plans do not
  include the expected increase in overhead by 30,
  which pushes the overhead well above the reserve.
  
• Recommendation Make presentation of what is
  unfunded explicit.
• Minor point Use of Access and Project does not
  yet (apparently) include resource loading. These
  details are being held at lower levels of project
  (software, tier1, etc.).

4
Management (2)

• (Formal) Change Control remains for the most part
  unimplemented, and is not being used as a
  management tool.
• The one change which exercised the mechanism this
  year was an aggregate omnibus change, which is
  not a recommended approach
• The move to PanDA should have invoked change
  control under any reasonable definitions of
  triggers
• The re-organization of the WBS should likewise
  have invoked change control, even if cost neutral
• A 3 month milestone delay as a trigger is much
  too lax at this stage of the project
• Changes in scope / cost as a trigger is too
  vague recommend adding an FTE-month or
  qualification
• Recommendation Utilize Change Control more
  formally.

5
Software (1)

• Data analysis model has matured to the level of
  having a concrete definition of the AOD
  (additional work needed is recognized)
• Further effort on optimizing read performance of
  the AOD ESD would be well spent now
• While the transient/persistent model shows good
  initial gains, should study whether simplifying
  the persistent data structures will in the end
  result in higher performance
• Geometry effort impressive, facilitating a nice
  event display

6
Software (2)

• Responded well to problems in DC2 and future
  scaling challenges by developing a tool (PanDA)
  that minimized reliance on the weak links in the
  grid toolset. Change made at good point early
  enough to do good job in hardening, not so early
  as to miss opportunity for maturing middleware.
• Presentations need polish PanDA might be viewed
  as an adhoc, non-collaborative effort. Need to
  emphasize that
• at top end, interfaces to standard executor
• at the bottom end, interfaces to standard grid

7
Software (3)

• Commissioning strategy...
• Component level testing for CSC is good
• Are the dates for milestones for components
  firm? Confirm they are early enough for
  integration testing.
• Delaying a full scale data challenge to actual
  beam may represent a risk

8
Software (3)

• Analysis Model needs additional work to show how
  pieces all connect together
• will the AOD be used, and give good performance?
• role of tag database (vs trigger data) isnt
  clear enough, relationship to streaming

9
Software Support

• Plans for software support are considerably more
  concrete (good!)
• How the 3 centers will be realized staffed
  remains vague
• Recommendation Identify people, charge them with
  the task, and get them to organize their support
  activities to accomplish the goals
• develop process that will motivate support, and
  not overload those tasked with it
• define / negotiate level of effort

10
Facilities

• Substantial progress on procurements and facility
  services (e.g. dCache, grid interfaces) and
  ramp-up of staff
• Wonderful news on WAN upgrades
• Capacity and staff are on schedule
• FY2006 will be another significant year of
  increases

11
Facilities (2)

• While current scale of dCache is impressive,
  current utilization is modest and does not yet
  represent a significant test
• Size of resources utilized through the grid
  interface averaged 100 cpus, which is rather
  modest need to work on larger stress tests

12
Facilities (3)

• BNL is enjoying significant benefit from
  co-location with RCF
• Need to document benefits, both to show the
  benefit to agencies, and to reveal potential risk
  if RCF goes away
• Budget does not include upcoming 30 increase in
  overhead.
• Recommendation include the overhead in out-year
  planning.

13
Facilities (4)

• Long term planning for Tier 1 (space, etc.) needs
  additional attention
• Current ESD size and simulation times are
  considerably above nominal, and may result in
  need for additional storage and compute
  resources, or a de-scoping of event rate.
• Recommendation Make sure this issue is being
  addressed at all levels within International Atlas

14
Facilities (5)

• Procurement of 2008 resources with fy2008 funds
  implies that resources are not fully available
  January 1.
• Recommendation
• Plans should be made to reflect the realities of
  procuring and commissioning multiple petabytes of
  disk and 1000 nodes (plan for staged roll-out
  during Jan-April 2008)

15
Grids

• Formal involvement in OSG management is on a good
  track
• expressions of support from US Atlas management,
  involvement of Torre
• There is some unease on both sides (OSG, US
  Atlas) about Atlas participation in OSG
• Recommendation Find success oriented approach
  (find a win-win) for US Atlas participation in
  OSG. OSG needs to deliver a net positive impact
  on the US LHC experiments early in LHC operations