280 likes | 292 Views
Fermilab staff collaborates in the CDF Experiment, providing support for upgrades/replacements, detector operation, and offline computing. Learn about the data logger replacement and the responsibilities of Fermilab in the CDF Experiment.
E N D
CDF ExperimentFermilab Experiment Support Stephan Lammel We, Sep26th, 2007
Overview • Fermilab staff from all around the Laboratory collaborate in the CDF Experiment: • Particle Physics Division, Computing Division, Accelerator Division and Technical Division • This brings a lot of expertise to the CDF Experiment • And gives the experiment access to a large variety of expert resources, • for consulting and in early stages of a project • experts to lead/support a project • testing/manufacturing places • contingency resources CDF, Fermilab Experiment Support, St.Lammel
Fermilab Experiment Support • Basic support as host laboratory: • installation, maintenance, technical, engineering, electronics, gas, networking, video-conferencing, safety, etc... • and beyond in especially three areas: • Upgrades/Replacements • Detector Operation • Offline Computing CDF, Fermilab Experiment Support, St.Lammel
Upgrades/Replacements • Data Logger Replacement • Detector Operation • Fermilab Responsibilities • Three Person Shift Crew • Remote Consumer Operator • Offline Computing • Fermilab Responsibilities • Expedited N-tuple Production • Glidein Processing Facilities CDF, Fermilab Experiment Support, St.Lammel
Data Logger Replacement • Original data logger for Run II was designed and built end of the 90’s for Run IIa luminosity by a team from Rochester and Tsukuba: • SGI/IRIX based, long term maintainability concern • single points of failure, redundancy not affordable • custom setup, loss of expertise concern • good but tight match to needs, little flexibility • data center buffer space, expensive to increase • complex, retro fitted to evolved data handling CDF, Fermilab Experiment Support, St.Lammel
Data Logger Replacement • Features of new data logger: • Use commodity hardware/solution where possible and appropriate • Share archival part of solution with D0 • Clean data handling interface, eliminate intermediate steps, collect and declare metadata directly for production use • Additional benefits of new data logger: • increased bandwidth to match DAQ/trigger bandwidth • 100 MB/s (from max 24 MB/s of original data logger) • Additional flexibility for high luminosity trigger environment • Increased buffer space and ability to archive data faster than to record them CDF, Fermilab Experiment Support, St.Lammel
Data Logger Replacement CDF, Fermilab Experiment Support, St.Lammel
Data Logger Replacement • Data logger replacement was a joint effort of Tsukuba, Rochester, Fermilab-PPD and Fermilab-CD • Fermilab-PPD brought in DAQ expertise • Fermilab-CD brought in metadata management and tape archival expertise • Tsukuba and Rochester brought in the expertise of the old data logger and did a lot of the integration work • under the leadership of Fermilab-PPD CDF, Fermilab Experiment Support, St.Lammel
Upgrades/Replacements • Data Logger Replacement • Detector Operation • Fermilab Responsibilities • Three Person Shift Crew • Remote Consumer Operator • Offline Computing • Fermilab Responsibilities • Expedited N-tuple Production • Glidein Processing Facilities CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Fermilab groups have agreed responsibilities for: • solenoid (sole responsibility) • silicon detector • COT (sole responsibility) • hadron calorimeter • muon systems • trigger (level-2 Pulsar, muon) • DAQ (sole responsibility) • consumer/monitoring CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Last year the operations department reviewed data taking, how to streamline operations, gain efficiency, etc. • The control room shifts were re-organized from four people to three of which one can be remote: • automate more common tasks to make the job easier and allow the shift to focus on critical decisions • transfer responsibilities to achieve more natural grouping and to better balance work load CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Three person shift crews started Dec 29th after several months of planning, review and preparation: • training of shift crew • computer console re-arrangement • automation of routine monitoring tasks • review of alarms and make selected ones audible • enhance interface to Tevatron • automate DAQ recovery for more frequent inhibits • two week trial run • Total savings of 4 FTE (3 shifts, 7 days/week) ! • Shift crew reduction was initiated and organized by Fermilab scientists CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Consumer operator, CO, monitors the data taking and quality. • He interacts with the scientific coordinator and detector/data taking expert but does not himself control the detector or data taking • With many collaborators in Europe and Asia participating in the detector operation from off-site is attractive to both the experiment and foreign institutions. CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Steps to realize a remote consumer operator: • The CO setup of the control room was duplicated in Pisa • All monitoring plots made web based by the consumer/monitoring team • IP based video link setup • A week long trial run with a remote CO and shadow CO in the control room in November • Since January one week-long remote owl shift per month from Pisa • Remote CO shifts from Tsukuba after the shutdown • Remote CO was initiated by a Fermilab scientists and setup together with a colleague in Pisa. CDF, Fermilab Experiment Support, St.Lammel
Detector Operation • Both the current operations heads are from Fermilab. • Three of the four associate heads are from Fermilab. • Half of the detector/online groups are lead by Fermilab people. CDF, Fermilab Experiment Support, St.Lammel
Upgrades/Replacements • Data Logger Replacement • Detector Operation • Fermilab Responsibilities • Three Person Shift Crew • Remote Consumer Operator • Offline Computing • Fermilab Responsibilities • Expedited N-tuple Production • Glidein Processing Facilities CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • Fermilab group has agreed responsibility for: • data handling, including metadata administration, tape archival and disk caches • full event reconstruction • central processing facilities • analysis framework and infrastructure • databases (software, interfaces, management) • offline operations • software release management/support • various reconstruction/analysis packages CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • The CDF analysis strategy is an evolution from earlier hadron collider experiments: • validate/tune reconstruction on small data subset • central store of reconstructed data • analysis through iterative selections • Although offline computing is based on commodity hardware/software, resources are nevertheless tight. • avoid duplication as much as possible • shared solutions to share support/maintenance • short latency/fast turn-around CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • CDF Computing Model (data flow view): • Raw data are written to tape (base of all physics) • Detector/data are calibrated (4-6 weeks effort) • Full reconstruction using on-site production facilities • Make highly-reduced analysis datasets, N-tuples • Monte Carlo generation/simulation/reconstruction • with run config matching data taking conditions • N-tuple selection for individual analyses needs • Further analysis on user desktop and at home institutions CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • CDF Computing Model (hardware view) • Enstore system provides IP-access tape storage • On-site processing facilities are used for: • detector/data calibration • full reconstruction • N-tuple making • analysis data selections (of the N-tuples) • Off-site processing facilities (CDF specific and Grid) • Monte Carlo generation/simulation/reconstruction • Disk cache of tape, disk pool, and project space hold the data analysis accessible CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • Data are grouped into few months periods by the offline to handle detector re-calibration needs: • datasets for calibration produced within ½ week • calibration plus validation within 3 weeks • full reconstruction finishes a month later CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • N-tuple (highly reduced analysis dataset) making issues and problems: • making N-tuples is very I/O intensive • it is non-trivial (need to learn N-tuple creation tricks) • physicists need to use the heavy analysis framework • making N-tuples takes quite a bit of CPU (latest analysis-level corrections are applied) • significant load on the central processing facilities • significant effort per analysis for each physicists CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • New N-tuple production: • common framework produces the 3 most used formats • shared expertise and resources • reduced load on on-site processing facility (N3) • fast, 3 day turn-around • off-loads N-tuple making from analyses • Expedited N-tuple production great success: • individual analyses start from much smaller production N-tuples • analyses are expedited • Fermilab group developed and manages expedited N-tuple production (from the need to reduce load on central processing facilities) CDF, Fermilab Experiment Support, St.Lammel
Offline Computing • Currently three on-site processing facilities: • production farm (does full event reconstruction) • “CAF” (executes user analysis jobs, CDF-only) • “FermiGrid” (latest user facility, unused cycles are being shared with other Fermilab experiments) • Working on evolving processing facilities into Grid based facilities with one on-site and one off-site job submission point for users: • idea and early development of glide-in mechanism by Fermilab-CDF person • development into common Grid tool by Fermilab-CMS CDF, Fermilab Experiment Support, St.Lammel
Example, Higgs Search • The search for the Higgs is the top priority for the CDF Experiment. • Additional sensitivity, beyond the increased integrated luminosity, needs to come from: • increased lepton acceptance • forward b-tagging • improved di-jet mass resolution CDF, Fermilab Experiment Support, St.Lammel
Example, Higgs Search • A Fermilab scientist is leading the effort to trigger on muons that fall into the phi gaps between wedges. • Forward, BMU muons have been pioneered by an ex-University of Wisconsin Fermilab scientist. • Fermilab postdoc modified tracking chamber digitization for increased hit efficiency at high luminosity. • A Fermilab scientist has reworked the tracking software for increased acceptance in the forward region and to get good efficiency at high luminosity: • forward b-tagging • lepton acceptance • A group to work and improve the di-jet mass resolution was setup by CDF last autumn. It is co-lead by a scientist of Fermilab. CDF, Fermilab Experiment Support, St.Lammel
Sumup • Fermilab contributes with 25 FTEs (of 65 total) to the detector operation. • Fermilab contributes with 12 FTEs (of 25 total) to the offline computing. (three of the five top leaders are Fermilab) • Of the top experiment leaders/managers 5 (of 7) are Fermilab scientists. CDF, Fermilab Experiment Support, St.Lammel
Summary • Fermilab continues to support the CDF experiment way beyond its host laboratory duties. • With upgrades winding down Fermilab scientist are looking for ways to increase the efficiency, shorten analysis turn-around, and reduce manpower. • Strong Fermilab participation in all main support areas of the experiment including collaboration leadership. CDF, Fermilab Experiment Support, St.Lammel