[the following is a kind-of consensus between HD/RHUL/SLAC or wishful
thinking that would provide benefits on the short time scale]
Hoi,
after a few phone calls, it seems that a new option is emerging.
Below you find a proposal, please let me know if you disagree. If I
hear nothing from you, I'll send this to vub-recoil after Easter.
The situation
-------------
o Currently, analysis-20 is being built and close to release. The
executable is expected to be about as fast as an OBJY exe ("a bit
slower"). It should be capable of running on OBJY micro.
o At the end of next week, a substantial amount of skimmed data is
expected to become available (as of now it's only 60/fb). By then
it might be something like 130/fb. Maybe/hopefully.
o Clare needs (or would like to use) more than just 80/fb for her
thesis.
o The same is more or less true for the b2ulnu analysis (and Ed's
thesis)
o The projection for skimmed MC CM2 availability is slipping. At
the Wednesday physics meeting we heard "June".
The idea
--------
o Produce new ntuples with minimal (=no) changes with respect to the
old "big" ntuples. Except: Just produce ROOT files. This allows to
use them for analysis immediately and is probably the only way to
use RUN3/4 data for public results before the end of this year.
This is explicitly NOT a CM2 analysis, just an attempt to get more
data for the short time scale.
o Run on RUN4 CM2 data as more becomes available
o Run on MC as soon as possible. We should test whether running on
SP5 OBJY is viable.
The plan
--------
o Ed will provide a set of tags to build IslBrecoilUserApp, based on
analysis-20. He will do basic validation, i.e. that it is running.
This also includes tcl (steering) files for CM2 and OBJY running.
o Urs will do a bit more of validation, looking at all variables in
the "h1" tree.
o Royal Holloway will organize the production. I think this will
involve the following:
- backup the current HBOOK ntuples of Henning/Oliver to mstore
and/or RAL or somewhere else.
- Create tcl files for skimmed data. In the following I detail
what issues must be considered in a low-tech approach (based
on "run", the run-script with built-in bookkeeping and
optimized queue saturation :-) Other possibilities exist of
course, I just don't know them. Whoever organizes the
production is free to choose whatever works!
+ The naming scheme for the "basename" should be well
designed. In the last production we had a bit of a mess and
it made life difficult. . A possible solution is something
like the following:
genbch-run1-.....
genbnu-run1-.....
genccb-run1-.....
genuds-run1-.....
cktbch-run1-.....
cktbnu-run1-.....
cktb2u-run1-.....
b2unre-run1-.....
b2ures-run1-.....
b2umix-run1-.....
NOTE: The total amount of files will likely exceed 100000.
(We had something like 30k for the previous production.)
NOTE: We used to have something like 2k events per file. We
have to think whether we want to merge the rootfiles to
reflect the merged CM2 files. Another possibility is to have
in the filename (in the ..... part above) the start and end
events in the merged CM2 files (see next item if this NOTE
is not clear).
+ The size of the tcl files needs to be optimized for the
queue length. (kanga?)
NOTE: I think this could mean that we cannot run one job per
CM2 merged skim file. This needs to be studied!!!
+ The tool of choice is probably "BbkDatasetTcl".
+ I think that the tcl files should be in a logical directory
structure to avoid too many files per directory
$BASE/tcl/SemiExclBreco-2004a/data
$BASE/tcl/SemiExclBreco-2004a/data/run1
$BASE/tcl/SemiExclBreco-2004a/data/run2
$BASE/tcl/SemiExclBreco-2004a/data/run3
$BASE/tcl/SemiExclBreco-2004a/data/run4
$BASE/tcl/SemiExclBreco-2004a/mc
$BASE/tcl/SemiExclBreco-2004a/mc/run1
$BASE/tcl/SemiExclBreco-2004a/mc/run1/bch
$BASE/tcl/SemiExclBreco-2004a/mc/run1/bnu
$BASE/tcl/SemiExclBreco-2004a/mc/run1/ccb
$BASE/tcl/SemiExclBreco-2004a/mc/run1/uds
$BASE/tcl/SemiExclBreco-2004a/mc/run1/sig
$BASE/tcl/SemiExclBreco-2004a/mc/run1/ckt
- The output root files should be stored in a way that reflects
this structure:
$BASE/output/SemiExclBreco-2004a/data
$BASE/output/SemiExclBreco-2004a/data/run1
$BASE/output/SemiExclBreco-2004a/data/run2
$BASE/output/SemiExclBreco-2004a/mc/run1
$BASE/output/SemiExclBreco-2004a/mc/run1/bch
- A few notes on the directories:
+ It does not really matter what names we choose, but it
should be something that is consistent and extensible to
new productions, which could end up in, e.g.
$BASE/SemiExclBreco-2004b/
+ We should avoid too many subdirectories, but should make
sure that not too many files end up in one directory.
(Note: In the old production, 80/fb data and 240/fb MC, we
had 11000 gen B+ files in total.)
+ Not all directories need to be physically below $BASE, they
could be symbolic links to a different disk. But we should
see all from one base location.
- Of course, the logfiles should be stored similarly:
$BASE/log/SemiExclBreco-2004a/data
$BASE/log/SemiExclBreco-2004a/mc
and the corresponding subdirectories. If we use "run" this is
essential.
- The jobs will be run by a bunch of people, organized (and
tabulated) by someone. "Volunteers" so far are
Clare
Ed
Henning
Rolf
Urs
Oliver ('s account, at least)
Other GradStudents
Given this amount of manpower, we might actually get through
the unskimmed SP5 OBJY (700/fb!) on a relatively(?) short
timescale.
o Diskspace might be sufficient once we delete the old HBOOK files
(from the previous production) and then ask for some more when
it's critical and we have enough momentum.
Cheers,
--U.
|