lhcbcomputing manpower requirements. disclaimer m in the absence of a manpower planning officer, all...
TRANSCRIPT
LHCbComputing
Manpower requirements
2
Disclaimer
m In the absence of a manpower planning officer, all FTE figures in the following slides are approximate
m In particular, there may be omissions (at the level of 0.5-1 FTEs) in the tables describing currently available effort
3
Long term computing project responsibilities and needs
o Project management (2 FTE)P Coordination, Planning (resources, activities,
development), Liaison with outside bodies (WLCG, RRB, LHC experiments)
o Software engineering support (4 FTE)P Code and release management, nightly builds, software
performance infrastructure, user environment, tutorials, documentation
o Central infrastructure support (1 FTE)P VO management, CERN-IT liaison, Web services, Vidyo
o Applications coordination, maintenance, integration (6 FTE)P Framework maintenance (Gaudi, Persistency, Event
model etc.)P Conditions database development, coordination,
deploymentP Physics applications release planning, integration,
performance and regression testing, validationd Gauss, Boole, Brunel, DaVinci, Moore, Event display etc.
o Computing operations (8 FTE)P Production planning, production management, data
management, grid operations, user supporto Distributed computing software maintenance (8 FTE)
P Dirac+Ganga coordination and integration, book-keeping, databases, production tools, monitoring, accounting
4
Manpower currently committed to core activities
Country FTE
Brazil 0.4
France 0.5
Germany 0.6
Italy 3.5
Russia 1.1
Spain 1.5
CERN 8
Switzerland 0.5
Netherlands 1.0
United Kingdom 5.0
United States 0.5
TOTAL 22.6 (c.f. 29 needed)
5
Current manpower
m Current manpower insufficient to cover core activitieso Estimate 29 FTE needed, 22.6 FTE available
P Some activities not covered (see next slide)
m Very little manpower available for non-core activitieso ~4 FTE at CERN in principle working on Gaudi and
Dirac software developmentP In practice making up some of above missing manpower
o Small pockets of effort in various countries, for example:P Spain (DIRAC development)P Italy, UK, CERN (Data Preservation and Outreach)P Italy, Netherlands (Multicore R&D)
m Barely sufficient to keep our software and computing abreast with evolving technology
6
New activities
m Core activities not covered by existing manpowero e.g. documentation, tutorials, event display,
software validation, performance and regression testing
m Software improvement activities for upgrade conditionso Application software development
P e.g. Coordination of GPU activities, frameworks for multicore, adoption of Root6.
o Software optimisationP e.g. vectorisation, architecture dependent compilation,
C++11o Data Management
P e.g. Use of data federations, data popularity, remote access to data, event indices, optimisation of Root I/O
o Distributed ComputingP e.g. Virtualisation, Interfaces to Clouds, Multicore
queues, DIRAC scalabilitym Data preservation and open access
m Preliminary estimate: a further 10 FTEs needed
7
What do other experiments do?
m Atlas, CMS, Alice all have some core activities covered by M&O A (either cash or in-kind manpower contribution)
P Software engineering supportP Central productions and operationP Central infrastructure support
o Atlas + CMS: ~2 MCHF (or ~20 FTE, largely “in kind”)o Alice: 0.5 MCHFo (LHCb: 170 kCHF for subsistence)
m In addition:o Atlas itemises all computing contributions under
M&O BP 171 FTEs in 2013
o CMS finances additional core computing manpower at CERN through M&O BP 8 FTEs
m All have formal agreements of where manpower comes from.
8
Observations
m Manpower currently devoted to operations is incompressibleo Compares very favourably with situation in GPDs
P BUT many tasks do not scale with collaboration sizeP AND data handling for LHCb in upgrade comparable to
GDPs in Run 1
m Funding for computing resources is (at best) following a constant budgeto Growth per CHF follows Moore’s Law only if the
software is optimised for new architectureso Growth of LHCb requirements is steeper than
Moore’s law
m Major evolution of computing model and software requiredo Requires significant injection of new manpower
P Initially for coordination and R&DP Subsequently for deployment and operations
9
Possible scenario
m Divide computing project into a number of work packageso Each including organisational, development and
support componentso Work in progress
m Ask individual groups (or countries) to volunteer responsibility for one or more work packageso Each contribution: team of several people
P Size will depend on work package, but 1-2 FTEs will be a minimum viable contribution
o Similar to sub-detector organisationP Resulting in a document describing sharing of
responsibilitieso Precise sharing of responsibilities is essential
P Best effort like now is not enough for the upgradem Add up contributions. If large shortfall, may need
to introduce charging for computingo e,g. funding of core software services through M&O
A contributions, either in-kind or through a ‘tax’m Principle to be discussed in CB this week
10
Other news
m Ricardo Graciani’s mandate as computing resources coordinator is overo Concezio Bozzi (Ferrara) has agreed to take on this
reponsibility