Transcript Trigger Requirements - Universita' del Salento

Trigger Requirements
first thoughts from
long list to
acknowledge
people who
reacted
on a very
short notice
Alessandro Di Mattia,
Giovanni Siragusa,
Sergio Grancagnolo,
Andrea Ventura,
Michela Biglietti,
Diana Scannicchio
John Baines,
Dmitri Emelyianov,
Iwona Grabowska-Bold,
Alan Poppleton
LVL2 MS
EF MS
ID trigger / PESA
LVL2 ID
EF ID
ID/Muon fitting in offline
ID requirements
Current map field size and access time OK for ID algorithms in HLT
At LVL2 the access time is about 10% of the overall latency of the
algorithm BUT they would not like to see this increase.
Impact on performance of a similar size map, less symmetric and
less granular in B-filed steps needs to be studied.
Field / Pt precision:
> hard to quote an absolute number
> let’s put it in the other way: ID HLT needs the best resolution
possible with a field map size acceptable
EF
Hardware baseline setup of EF:
8 core PCs, 1GByte memory per core (upgradable to 2)
1 processing task (EF process) running per core
1 GByte of memory available per process
Recent Technical run:
exercise the HLT sw on pre-series of the final hardware
measured ~700 MBytes allocated by a single EF process (running EF for all
slices: single muon/electron/gamma/tau/etc… trigger) using standard
Athena Tools (i.e. magnetic field map and access tools)
B-field of up to 100 MBytes (now 30MByte) are affordable and leave
contingency for extra algorithms to be included in EF
1 GByte possible only on upgraded hardware (not foreseen at the moment)
EF Muon Spectrometer
TrigMoore uses –up to now- the offline iPatRecFitter (global
Runge-Kutta based fit)
Access to the field was optimized at some point on the available map: step
in most of the MS volume is 320 mm, sometimes 160, 80, 40 mm (3
accesses per step to account for the local gradient) – the optimization
was based on the process of using a very low step size as starting
condition and increasing it up to un-observable pt measurement
degradation (both on average value and resolution);
Depending on the region, a given step is chosen;
 Without re-optimizing the access to the map (the step size in each
region) the size of the magnetic field map does not impact on the
number of accesses but just for the time of a single access takes;
 Field look-ups are probably the timing bottleneck for the fit !
 MooMakeTracks (fit in TrigMoore based on iPatRec) takes more than 1/3
of the total TrigMoore time/event: 400ms for single muons in nominal
background conditions at high luminosity) – 1s EF total latency
From M.Biglietti and TrigMoore group talk
at T&P week in May 2006: Muon EF performance
EF Muon Spectrometer
TrigMoore performance
Long study to see the effect of a coarse grained map used in EF
reconstruction for a detailed (nominal) map in simulation or for data !
The exercise just started with the following MS maps:
• default map
• 40gaus step for MS only
• 80gaus step for MS only
• 80gaus step + doubled step in phi and z (MS only)
look at total timing of TrigMoore/track (not even isolating the fit !)
– no observable effects
look at measured pt for 6 GeV and 20 GeV single muons (integrating in
eta/phi !) granularity in phi-eta of the performance study would be
necessary  high statistics needed
– no observable effects
EF Muon Spectrometer
Def. Map
40 gaus step
80 gaus step
80 gaus step
+double step
in phi/z
ptrec/pt
6 GeV
sigma
%
ptrec/pt
20 GeV
sigma
%
1.004±0.002
4.1 ± 0.2
1.002±0.001
1.004±0.002
4.1 ± 0.2
1.001±0.001
3.0± 0.1
mean 70
rms 158
1.006±0.002
4.3 ± 0.3
1.002±0.001
3.0± 0.1
mean 83
rms 166
1.008±0.002
3.8 ± 0.2
1.000±0.001
3.0± 0.1
mean 72
rms 171
2.9± 0.1
TrigMoore
time/track
mean 79
rms 166
Use of magnetic field in the LVL2
Muon Trigger
Use of magnetic field map of Toroids in the LVL2 Muon
Spectrometer Reconstruction for:
•
•
filling the performance gap that endcap Reco has with respect the Barrel
Reco: the LUT approach is swamped by the high inhomogeneity field;
get rid of part of the resolution tails, which give efficiency problems while for
tagging high-pT muons and rejecting low-pT muons.
Goal: build a kalman filter procedure that solves the left-right ambiguity of the
MDT drift circles while prividing a global fit of the muon track.
Timescale: work should hopefully start this september.
Target accuracy of the map of Toroids:
•
•
having a pT/p resolution from fit of about 4% is good for the LVL2 MS
reconstruction;
but this must match the online requirements (next slide);
Online requirements @ LVL2
Map size: the limit on the size can be set up to 100 MByte.
– LVL2 farm will employ 8 core processor machines equipped with
8 GByte of memory;
– almost 8 L2PU instances will be run, each taking ~650 MBytes (
data of the last technical run at point one with the combined
LVL2 trigger slice);
– this would allows for adding 1 or 2 more L2PUs instances in
order to optimize the use of CPU.
Access time: goal is to limit the overall access time below 1
ms.
– access time of 1 ms per cell is fine, assuming a maximum
number of 700 tracking steps;
– the overall time increase due to the fit is expected to be ~1.5 ms
and in any case tunable runtime acting on its precision.