Gaudi Developers Meeting - 2015-06-10
=====================================

ATLAS
-----

Charles - all we have done so far is the forward scheduler with data
dependencies. Not yet tried graph scheduler.

Why use MP at all? Charles - calo testbed was limited with some serial
algorithms, so muliti-process allowed us to use a different strategy
for more concurrency, but to benfit from CoW for memory
savings. Shares more memory than running multiple jobs in parallel.

Was this forced by technical/software limitation? Charles - memory
consumption is a primary driver, so this was a useful thing to
do. Balance memory usage against performance.


Ilya - with Marco have started to work on efficient scheduling of i/o
bound problems.

Pere - I/O is mainly a ROOT issue. Are working on this in ROOT dev.


Benedikt - Need to agree on the future of Gaudi. Can we sync on the
handle useage and not do things twice?


ACTION: Graeme to forward details of ATLAS frameworks workshop to
wider lists.



FCC
---

Marco - Would extended features become too bloated? Some of these
decisions need to be experiment specific? Propose a generic base, but
can be specialised...? Note too many particle property services already!

Ben - small core and modules on top?

Paolo - "contrib" addons.

(Importantly we have a component model.)

FCC was able to use Gauss with very few changes - good generic
example.

Should have a turnkey solution for building and installing Gaudi.

Marco - would be nice to have it buildable on system libraries. Much
easier for users.

Should consider deployment as well.


LHCb
----

ATLAS happy to have disruptive changes in long term as long as Run2
stability is assured.

Paolo - licenses are really difficult and time consuming... can
prevent people from contributing.

Paolo - would like to have much more agile development with
Gaudi. Work more continuously.

Marco - git(lab) offers an easier way to review and merge
contributions.

Paolo - can we be more adventurous with patches in the future? i.e.,
not too formal. Seb - this can be lightweight. Paolo - lack manpower
for tests.

Benedikt - we need to invest in proper setup so everyone can run the
tests.

Andrea - what is the test status in Gaudi? Marco - I try to write tests
for every bug report. Problems come from the fact that Gaudi isn't
easily testable.

Rare bugs can be hard to test - but code can be fixed. Static checkers
would help a lot.

Andrea - effort to write tests is not wasted.


Pere: SFT would like to contribute some infrastructure support (build
and test suites, etc.).

ACTION: Pere and Benedikt to look at infrastructure support
possibilities from SFT.

Benedikt: Clean up interfaces.


Proposed: Adopt a bi-weekly meeting schedule.

Proposed: Marco, Benedikt and Charles to take a 'interim leadership'
role and organise the future meetings and topics.

* Both accepted by the meeting.



Identified Future Meeting Topics:

- Concurrent I/O
- Data Handles
- Use of modern C++
- ATLAS Future Frameworks Report