L0 implementation recommandation (2nd try) (lhcb-trig)
Time: 2:29:04 PM
Remote Name: 22.214.171.124
Yesterday something went wrong with my mail, hence only
email@example.com got it, and not firstname.lastname@example.org. However, all already got a
copy of last year requirements list. This is the 2nd try of the implementation
This memorandum intents to compare the various proposal for implementing the LHCb-L0
calorimeter and muon triggers. A short history: at the end of 1998 a requirements list was
circulated (included as an attachment to this mail), and the groups were asked to submit
written proposals addressing these requirements before April 1, 1999. No group objected to
To discuss the various notes and give feedback to the authors the following
committee was set-up: (of which some as indicated in the requirements document were asked
to pay special attention to specific items): Frank Harris, Ioana Videau, Jorgen
Christiansen, Burkhard Schmidt, Ivan Korolko, Renaud Le Gac, Yuri Ermolin and Hans
Dijkstra. This group met several times to discuss and comment on the proposals received.
On May 7 the following detailed time-table was distributed to the groups:
- all info must be received before May 14. This must be the real dead-line!
- May 20, 9:30h: expert group formulates last questions to submitted papers, 1 week for authors to answers questions,
final info ready May 28.
- June 1: expert group makes a "consumer guide" to compare the proposals with the requirements, and will formulate a
consensus proposal for Technical Board(TB) if possible. This will be mailed to all participants for discussion
before the LHCb week.
- LHCb week June: discussion on consumer guide (TB, trigger meeting,
plenary) and "hopefully" consensus proposal to TB on Friday).
Friday of LHCb week, present this to TB if anything changed from present memorandum in the LHCb week.
The following notes were evaluated (and references therein):
Note Received "my" nickname LHCb 99-004: (Feb 17): 3D flow proposal.
LHCb 99-007: (Apr 26): 2x2 implementation
LHCb 99-008: (Apr 02): muon processor from Marseille
LHCb 99-013 (May 06): cal-trigger Bologna
On May 3 Dario Crosetto informed us about note LHCb 99-006, but it was decided to postpone
its reviewing since no sufficient answers were yet received to the
questions posed on March 15 about note 99-004. A reaction to these questions was received
on June 2, too late to be included in the considerations presented below. However, most of
the arguments in the consumer guide apply to the architecture, and are not expected to
change. In the 3D flow proposal the application to the muon trigger is hardly addressed,
however it is felt that the outcome from the consumer guide will not change if applied to
the muon trigger. Victor Golovtsov promised to try to submit the Gatchina approach for the
muon trigger with the aim to contribute ideas to another group in LHCb No paper has been
submitted, but several presentation were given during past LHCb weeks. The 2x2 (Orsay) and
Bologna groups have decided to submit a combined proposal, "based on the best parts
of the individual proposals". This proposal is not ready yet, however some ideas have
already slipped into the final version of the Bologna note. This memorandum will first
list how the ad hoc committee mentioned above evaluates the different proposals with
respect to the requirements list. For this purpose a "consumer guide" approach
has been taken, attributing ++/+/+-/-/-- as a rough indication on a scale ranging from:
completely satisfactory to absolutely insufficient. It should be emphasized that the
requirements do not all have equal weights, and not all items are addressed. However, they
should be considered as the basis for a decision on the type of implementation we would
like to pursue for writing our TDRs which should be ready by mid 2001.
requirement # 1 2 3 4 5 6 7
3D + - +- + - ++ --
2x2 ++ ++ +- + +- + ++
Bologna ++ + +- + +- -+ ++
Marseille ++ ++ + + + ++ ++
Comments per requirement:
1) Due to a lack of liaison with the FE groups, the 3D implementation would not fit the present FE
proposals. The solution seems to push the connectivity to the limit.
2) For 3D no clear idea is given about the total cost. Bologna has probably slightly underestimated the total
cost, and is quite a bit more expensive than 2x2.
3) 3D has a variable decision latency, the latency depends on the history of the events. Some of the consequences have been
simulated, however not with realistic MC data. In the 2x2 and Bologna proposals the very
last steps do not have a fixed latemcy. No simulation of the consequences is presented,
apart from heuristic arguments. For the combined proposal mentioned above it is claimed
that the calorimeter implementation is fully synchronous. Marseille has simulated the
consequences of the non synchronous implementation. This study is not complete, partly due
to a lack of centrally generated data. Indeed if all conditions are pushed to their worst
to be expected simultaneously, the performance will degrade, however this is true for any
algorithm which could be envisaged right now.
4) All latencies, given the above disclaimers, seem to have been thoroughly estimated, especially for the Marseille option
given their simulation of realistic MC data.
5) The 3D has a BX number to compare how the data flows through the pyramid, but not clear how to handle/recover from a missing clock,
which might corrupt the whole system until a reset. 2x2 and Bologna will recover
automatically in their pipelined part of the algorithm but the data transmission to the L0
Decision Unit is not solved by these implementations and shifted to the L0-DU box.
(However, in the coming combined proposal this will be solved since the whole trigger is
claimed to be synchronous). The checksum bit as proposed by Marseille could maybe be
adopted by other L0 implementations as a light system allowing de-synchronisation checks
throughout the whole system.
6) 3D: can implement any algorithm (within limit of the depth of the pyramid).
2x2: changed TP algorithm, but thoroughly checked this in simulation, and found no significant
loss in efficiency However, consequences of luminosity etc.. not checked yet.
Bologna: modified algorithm, but not checked, modification will certainly not give an improved
efficiency. However, verified on real HERA-B data 2x2 and 3x3 give equivalent
performances, confirming the 2x2 simulation result in real data. In principle the Bologna
proposal could execute several algorithms, with the possibility for upgrades during the
Marseille: performed simulation addressing all points.
7) As mentioned above: the groups from Orsay and Bologna decided to join forces, and in their proposals already
indicated what R&D they for-see up to the TDR. 3D flow has no viable support from any
group, nor a R&D program to follow up to the TDR. Marseille listed the steps up to the
The following Full Time Equivalents could be provided by the groups:
FTE engineers physicist technicians
Orsay: 2 2 (No FE overlap, dedicated to trigger)
Bologna: 1 2 2
Marseille: 2 2
All groups should be congratulated on the amount of work which went into
the proposals. . The ad hoc committee feels that given enough money and manpower, probably
all proposals could be made to work eventually, all be it that a clear advantage in
working fully synchronous is favoured to avoid variable latencies depending on
experimental conditions. . It should be made clear that these proposals are not the TDRs!
The aim is to select a base-line solution to prepare the TDR for a trigger implementation
based on the present proposals. .
The committee unanimously recommends the following proposals as base-line implementations:
- for the calorimeter triggers: the combined Orsay/Bologna proposal.
- for the muon trigger: the Marseille proposal.
Regards, Hans Dijkstra.