LHCb Core Software Meeting

Date and Location

03-Mar-2010
10:30 - 12:30
CERN (2-R-030)

Attendees

Gloria, Illya, Joel, Juan, Karol, Marco Ca., Marco Cl. (minutes), Markus, Paloma, Philippe, Rob, Silvia

Subjects

Round Table

Marco Cl.

  • There has been a discussion recently on the limitation that we cannot have two instances of the same ConfigurableUser class.
    • a small group of people will try to discuss the problem
    • a preliminary fix is available in Gaudi since yesterday afternoon

Marco Ca.: this is not a minor change and we should carefully validate it before putting it in production.

Markus

Marco Ca.: There has been a lot of fuss in the Online about the quantity of histograms in Brunel. A fix is available in the latest release.
Gloria: Currently, Brunel is disabled in the monitoring farm because of the problem. It should be re-enabled.

Karol

Marco Ca.: Which is the status of the nightly builds? Windows?

There build on Windows work only when run by hand (i.e. not through "Scheduled Tasks"). Marco Cl. suspects of a fake failure due to the conditions of execution.

Plan to change the implementation of the build procedure (better integration in the LCG system).

Joel

  • The new lxbuild machine is on its way.
  • Discussed with Patrick about the SAM test of DaVinci: it should not take 1.5 hous.
    Juan will fix it.
  • In the production request, there are fields for projects and specific packages, but the version of SQLDDDB cannot be specified.
    We must ha the possibility to select if to use Oracle or SQLite and, if SQLite, which version of SQLDDDB.

Marco Ca.: there is a fix for the SAM test in the latest version of Brunel.

Silvia

  • Testing the new way of configuring the Geatn4 physics list constructors

Gloria

  • Magnus has observed a problem in Gauss when generating rare events. If too many events are generated and thrown away, the application stop without messages. Investigations ongoing.
  • Warning in the nightlies about the no-pyzip tag.
    Marco Cl. says that it happens if the package doesn't use GaudiPolicy.

Marco Ca.

  • Last week has been discovered by chance that the decodings of the raw banks were not checking the MagicPattern. Task #14202 has been created and the code is being fixed.
    Marco Cl. pointed out that the sanity check should have been put in the framework (since it doesn't depend on the detector) instead of in each decoding... but it is too late to do it now.
  • The bookkeeping should return the configuration for the Conditions Database tags to use together with the files...

(...long discussion...)

It has been agree that the default tag to use should be taken from the data and the system (UpdateManagerSvc, etc.) should be able to switch from one tag to another between two events (see task #14262).

Rob

* The Integration project is getting ready and should be soon introduced in the nightly build slots

Juan

* There is a problem with a deprecation warning in some DaVinci Python code spot by Rob's test on the instantiation of Configurables. To be fixed.

Results on ROOT I/O investigation (Markus Frank)

(slides)

Markus presented some preliminary results on the comparison of I/O performances in POOL (several TTree with one branch each) and with the ROOT suggested way (one TTree with several top branches).

It seems clear that, in the tested case, the POOL approach is not much slower than the ROOT one for local files. It suffers when using the "castor" protocol (rootd), where the optimizations in ROOT (fewer round-trips) allow a 40% better speed.

More tests with more realistic scenarios will be done. For the moment, it is not extremely clear how much we are I/O bound and if the gain of 40% there is significant in a normal job.

-- MarcoClemencic - 03-Mar-2010

Edit | Attach | Watch | Print version | History: r5 < r4 < r3 < r2 < r1 | Backlinks | Raw View | WYSIWYG | More topic actions
Topic revision: r5 - 2010-03-04 - MarcoClemencic
 
    • Cern Search Icon Cern Search
    • TWiki Search Icon TWiki Search
    • Google Search Icon Google Search

    LHCb All webs login

This site is powered by the TWiki collaboration platform Powered by PerlCopyright & 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback