420 likes | 559 Views
WLCG INFNGRID Tier1 Tier2. Relazione referee calcolo LHC. Francesco Forti, Università e INFN – Pisa Per il gruppo di referaggio: F. Bossi, C. Bozzi, R. Carlin, R. Ferrari, F.F., M.Morandin, M. Taiuti. LCG Comprehensive Review. LHCC di Settembre 2006
E N D
WLCG INFNGRID Tier1 Tier2 Relazione referee calcolo LHC Francesco Forti, Università e INFN – Pisa Per il gruppo di referaggio: F. Bossi, C. Bozzi, R. Carlin, R. Ferrari, F.F., M.Morandin, M. Taiuti
LCG Comprehensive Review • LHCC di Settembre 2006 • Due giorni di review di LCG LCG Phase 2 = WLCG • Presentazioni dei referees: F.Forti - Referee Calcolo LHC
10 k WLCG Infrastructure • Based on two major science grid infrastructures: • EGEE (Enabling Grids for E-science) • phase 2 approved after last CR, funded until Apr 2008 • OSG (Open Science Grid) • 5-year funding cycle pending approval with DOE/NSF, (positive) decision expected in a few months • At time of 2005 CR, interoperability between grids was a major concern • this issue has been worked on in the meantime: authentication, job submission, mass storage access across grids show progress, though no common interface in sight Jobs per day (EGEE grid) Jobs per day (OSG grid) F.Forti - Referee Calcolo LHC
Metrics & Monitoring • Monitoring of availability & reliability has been major milestone • for T-1 centers now done now regularly (fails on some sites) • still below MoU level ~74% • Monitoring of job failures at application level is much harder • experiment dashboards • analysis of job logs. Still much manual work. • reliable automated system for job failure classification not around the corner • key point to sustained reliability should be pursued with priority F.Forti - Referee Calcolo LHC
Accounting CERN + T-1s • Since ~4 months, full accounting data for CERN + T-1s • comparison with installed & pledged resources • Monthly use relatively low • related to present use pattern (testing/ commissioning/ challenges) • No indication that performance bottlenecks may be due to resource limitations F.Forti - Referee Calcolo LHC
Impact of Schedule Change • Reminder: running scenarios assumed for TDR requirements: • 50 days of physics in 2007 • 107s pp + 106s AA in subsequent years • New scenario after revision of schedule: • Experiments will provide revised estimated requirements by begin of October WLCG & funding agencies • preliminary (non-endorsed) numbers exist from ALICE, ATLAS & LHCb NOTA: il processo di revisione è ancora in corso F.Forti - Referee Calcolo LHC
Shortfall of 13.9 MCHF for phase 2 (as of Apr 2006) reduced to 3.4 MCHF • Assuming the preliminary numbers (!) from the experiments’ revised requirements estimate F.Forti - Referee Calcolo LHC
WLCG Personnel • Much depends on a suitable succession project to EGEE-II from Apr 2008 onwards • 15 FTEs at stake alone at CERN • similarly crucial for external centers • This is a point of concern. WLCG should strive for a consolidation in terms of a more structural project, in particular also at the level beyond T-0 F.Forti - Referee Calcolo LHC
Commissioning Schedule • Still an ambitious programme ahead • Timely testing of full data chain from DAQ to T-2 chain was major item from last CR • DAQ T-0 still largely untested F.Forti - Referee Calcolo LHC
Middleware • Very significant progress during the last year on middleware and grid activities by the different experiments. • A system is in place and works in scheduled production periods. • It has been used by the experiments and if/when stable and reliable it should meet needs. Now robustness and stability is the key to make sure the system survives heavy (unscheduled) use as LHC startup approaches. • Many important aspects still not totally accomplished (remote site monitoring, accounting, job priorities & user tools) essential in a realistic system for a running experiment • Fundamental to allocate the required level of manpower beyond 2008 to maintain basic functionality, user support, upgrades and interoperability among grids. • Interoperability essential to make use of all available resources F.Forti - Referee Calcolo LHC
EEGE Middleware Development • gLite 3.0 • Successfully deployed in May 2006 • Debug of different components still continuing • Reliability, reliability, reliability • 50% resources spent on user support of existing infrastructure and software bug fixing. • Current activities (triggered by experiments) • Security • Data Management • Usage Accounting • Job Priorities (new GP-Box project…one year time scale) • Job priorities: absolutely non trivial when it is a decentralized system…experiments should carefully develop and manage this (perhaps starting from existing examples in running experiments) F.Forti - Referee Calcolo LHC
Application Area Projects • SPI – Software process infrastructure (A. Pfeiffer) • Software and development services: external libraries, savannah, software distribution, support for build, test, QA, etc. • ROOT – Core Libraries and Services (R. Brun) • Foundation class libraries, math libraries, framework services, dictionaries, scripting, GUI, graphics, SEAL libraries, etc. • POOL – Persistency Framework (D. Duellmann) • Storage manager, file catalogs, event collections, relational access layer, conditions database, etc. • SIMU - Simulation project (G. Cosmo) • Simulation framework, physics validation studies, MC event generators, Garfield, participation in Geant4, Fluka. F.Forti - Referee Calcolo LHC
AA Example - PROOF Relative speed-up F.Forti - Referee Calcolo LHC
AA – CONCLUSIVE REMARKS 1/2 • Lots of of work • important progress and achievements • Managerial difficulties due to the project fragmentation • the Simulation project • some difficulties in interfacing some Monte Carlo generators to the LCG simulation infrastructure • ROOT project • properly managed ; appropriate manpower resources • achievements: consolidation, fast access to data • Merging of SEAL (Shared Environment for Applications at LHC) progressing successfully • important progress of PROOF, powerful tool to extend ROOT to run on a distributed, heterogeneous system • Alice, CMS and LHCb are expressing interest in using of PROOF • clear decisions by the experiments needed F.Forti - Referee Calcolo LHC
AA – CONCLUSIVE REMARKS 2/2 • persistency framework project • key ingredient for LHC computing • difficult to assess the progress level • important effort by AA to keep the link with the experiments and the users strong and effective • LCG Generator monthly meetings • Architects Forum, AA Meetings every 2 weeks • Savannah portal • manpower • present level globally very near to the needs • some reassignment can cure the limitations affecting individual projects • possible manpower crisis in 2008 (retirements and contract ends) • appropriate action be taken in2007 to guarantee adequate manpower level in 2008 and beyond F.Forti - Referee Calcolo LHC
Computing fabric - CERN • T0 and CAF are well on track • But aggregate capacity for 4 experiments not demonstrated • Still slightly underfunded despite recent improvements • Impressice empty space in computer center • Building, cooling and power upgrades planned as required • T0 well understood • Demonstrated capabilities in full scale ATLAS test • CAF requirements still not well defined • CERN Analysis Facility or Calibration and Alignment Facility ? • Experiments need to deliver well in advance • Keep in mind purchasing cycles of 6+ months • Storage systems have improved performance • Still adding features, need ongoing attention • Manpower tight; need perspective with EGEE successor • Scalability - still an order of magnitude to go: • CASTOR2 and Directory service are critical F.Forti - Referee Calcolo LHC
Computing fabric – storage • Storage Resource Manager v2.2 • WLCG Standard storage interface, defined in May 2006 • Hybrid between 2.1 and 3.0 • Implementation is essential for LCG service • Castor2 • Deployment at T0 successful, well integrated • Inherent performance problems hit ATLAS and CMS, fix underway • Tier sites had problems - high support load for CERN • Review in June positive towards the project, but • „Many years of […] periods of operational distress“ • dCache • Project manpower has improved - 1 FTE for dCache user support now • No clear deadline for implementing SRM v2.2 - But seems to be on track • Community Support: OSG fund their own requests • DPM – Disk Pool Manager • In widespread use at 50+ smaller sites • Will be late in implementing SRM v 2.2 • Serious manpower troublesNot an issue for T0 and CAF • Indirect issue for T1s (transfer to/from T2s with DPM) F.Forti - Referee Calcolo LHC
Fabric – Distributed Deployment of Databases – 3D • Provides database infrastructure and replication • eg. for detector conditions and geometry, file catalogues, event tags • Initially set up at CERN and 6 “phase 1” Tier1 sites • to do: monitoring and (at some sites) backup systems • Replication performance sufficient for conditions and catalogues • T0->T1 replication 50% of T0->T0 rates. More optimisation possible • Moving from R&D to service phase • Experiment data rates and access patterns still not fully known • All experiments are testing real applications on the real infrastructure • Tier1 resources should be adequate • CNAF one of the first sites online for 3D F.Forti - Referee Calcolo LHC
TIER1 F.Forti - Referee Calcolo LHC
Tier1 Issues and recommendations… • Tier1s need to know the consequences of the schedule change • required resource changes affect procurement • Tier1s must be fully integrated in the experiments’ planning and decision process • Communication with experiments in vital to bridge the “culture gap” • Recommend liaison officer in both Tier1 and experiments • Meet regularly • Tier1 liaisons should attend experiment computing meetings • Experiment monitoring should be available to Tier1s F.Forti - Referee Calcolo LHC
…Tier1 Issues and recommendations • 24x7 operation • Requires “on-call” service • Still not at all Tier1s • Can never have all experts on call all the time • Not all problems resolvable by on-call responsibles • Can reduce outages, but some are still unavoidable • Coordinate with experiments to avoid scheduling outages at multiple Tier1s at the same time • especially an issue with core MW upgrades • Stability of Middleware is crucial • Both problems and upgrades lead to down-time • Developers need to concentrate on reliability over functionality, and very well-tested releases F.Forti - Referee Calcolo LHC
7x24 Operations (K.Woller‘s view) • There´s no way to have experts 24x7 (x52) • Need to design services to survive trivial failures • Commercially available load balancers may help • Need to design for increased reaction times • By building service level redundancy where possible • For rare complex problems, „on duty“ coordinator may help getting the required experts together fast. What we have What people suggest 1 FTE 230 day x 8 hours 1840 h/year 24/7 expert service 8760 h/year 4.8 FTE F.Forti - Referee Calcolo LHC
Tier2 Summary • Tier2s are a very diverse bunch • 400-2500 CPUs, 50-800 TB, 1-4 experiments (also non LHC) • 1-13 staff FTE (most ~5), Mostly 1GB/s network, and no MSS (tape) • Most Tier2s participated in SC4 - Critical for experiments • Funding uncertainties • Some Tier2s are federations • up to 8 geographical sites • Mostly 1 CE/SE per site (ie. Middleware sees them as separate) • Share experience and some services, allow small sites to participate • Can work well, but requires close cooperation • Collaboration with “local” Tier1 is essential • Data transfers • Tier1 can provide advice and perhaps some services • CMS Europe: Not enough Tier1s for all Tier2s • 2/3 of Tier2s rely on DPM • concern for support and future compatibility (eg. SRM 2.2) • DPM support team is undermanned F.Forti - Referee Calcolo LHC
Old target – 1.6GB/s New target – 1.3GB/s Service Challenge • Old data, new limit • Come l’atrazina F.Forti - Referee Calcolo LHC
SC4 - What were the problems ? • No simple answer • Many, many individual one-off problems were mentioned • Little quantitative information was presented • Many reports of instabilities • T1 sites (ATLAS report all 9 T1s only all available for a few hours/month) • Hardware failures • SRM/mass storage • Castor/dCache • File catalogues • Site differences • Firewalls • Badly configured nodes/sites • EGEE software • File access (GFAL) • File transfer (FTS) F.Forti - Referee Calcolo LHC
SC4 – How to improve ? • Many comments that manual intervention was required • “heroic efforts” ; “at the limit of what the system can do” • Need for communication improvements and problem reporting between the sites • Error reporting, tutorials, phone meetings, workshops, Wikis, etc. • He sees this as the way to improve performance and reliability • Have to live with this level of problems; just get more efficient at overcoming them when they occur • Castor is a notable exception • However, must also put a lot of effort into bug fixing • Not “sexy”; may need to push to keep the effort in the right direction • Effectively division of effort in maintenance vs. development • Important to get the balance of effort right here F.Forti - Referee Calcolo LHC
SC4 – Other Comments • Experiments will not ramp up to nominal rates by Jul07 • E.g. ATLAS simulation is x10 below right now • Most are aiming for this around early 2008 • No direct DAQ output has been included yet • Hence, service commissioning period will not be based on realistic loads • Should commissioning targets be relaxed for 2007, given LHC schedule? Only makes sense if frees up effort to use elsewhere; not clear if true • Almost all service performance reported as data transfer rates • Obviously critical to get data out, both for storage and analysis • Some information given on job performance • Very little on CPU usage efficiency; this seems to be underutilised • Scheduled outages can be worse than unscheduled ones • They hit more than one site simultaneously • More than one item tends to be removed from service • A usable albeit imperfect service F.Forti - Referee Calcolo LHC
Coordination and communication F.Forti - Referee Calcolo LHC
Service coordination • Meeting structure setup to ensure communication • Make sure experiment and sites representatives have enough authority • Service Coordination Meetings ideally should regularly be held at each Tier1 site in addition to CERN. • Clear need for a sort of service operation coordinator that acts as a central collection point for everything’s that going on • Discussion on length of term for operation coordinator appointment • Should be reasonably long (>2-3 months) • Need to continue to increase the involvement of remote sites in the decision, planning, and monitoring process • Develop realistic plans and adhere to them • Convince remote sites that the plans are real • Keep everybody in the system consistenly informed • Careful in keeping the bureaucracy under control and the reporting load at acceptable levels F.Forti - Referee Calcolo LHC
Middleware/deployment/Service Challenges SUMMARY • Stability needs to be improved. No new functionality, but need stable, running service • Experiments need to start using all the features of gLite3, to find the new problems. • Need to keep developer to fix the bugs and make the system stable rather than devloping new nice functionality • Analysis of job failure rates still needs improvement • User support model needs to be revisited • maybe a first line of defense internal to the experiment • Target performance goals not quite reached • Continuous unattended operation still a long way off • A full scale test of the entire chain starting from experiments DAQ is still missing F.Forti - Referee Calcolo LHC
Fabric SUMMARY • Technologically there doesn’t seem to be an issue • Some scalability issues with LSF and service machines • CERN T0 still needs to demonstrate the full aggregated capacity for 4 experiments • CASTOR2 still an issue – critical item • Is manpower sufficient ? • Issue of external sites support • SRM 2.2 • Essential, but not yet ready nor deployed • dCache a bit late in developing srm2.2 • DPM – essential for small sites • Is it going to be supported in the long term ? If yes, need manpower. • 24x7 operation and staffing at external sites very difficult • Mixed level of readiness • To PROOF or not to PROOF • Encourage experiments to take a clear stand at whether they want it, since it has broad implications. F.Forti - Referee Calcolo LHC
Management and global issues SUMMARY • Involvement of external sites improved, but keep going. • Communication, communication, communication • Experiments involvement is essential • At CERN as well as at Tier1 sites • Staffing problem if there is not EGEE-III • How to make a transition to structural operation staffing • The modification in LHC schedules somewhat reduces the gap between needed and available resources. There should be no temptation for the funding agencies to reduce the level of funding. F.Forti - Referee Calcolo LHC
In Italia: INFN GRID • EGEE Operation: Il Regional Operation Center Italiano garantisce • Il funzionamento giornaliero della e-infrastruttura Europea • Un supporto per Molte VO (Virtual Organizations) sulla stessa infrastruttura multi-science • Sviluppo e mantenimento di MiddleWare: gLite • Garantire l’ evoluzione del Middleware Grid Open Source verso standards internazionali: OMII Europe • La disponibilità del MW in un’efficiente repository: ETICS • Partecipare alle attivita’ informatiche di Ricerca e Sviluppo • Coordinare l’espansione di EGEE nel mondo • Gilda – attività di disseminazione • EUMedGrid , Eu-IndiaGrid (MoU…) • EUChinaGrid (Argo..) , EELA(LHC-B…) • Sostenere l’allargamento di EGEE a nuove comunita’ scientifiche • GRIDCC (Applicazioni real time e controllo apparati) • BionfoGrid (Bionformatici; Coordinato dal CNR) • LIBI (MIUR; Bionfomatici in Italia) • Cyclops (Protezione Civile) • Garantire la sostenibilita’ futura delle e-Infrastrutture con consorzi et al. • A livello EU : EGEE II -> EGI • A livello Nazionale IGI • A livello di middleware EU OMII EU • A livello di middleware nazionale c-OMEGA • Coordinare la partecipazione all’Open Grid Forum (ex GGF) F.Forti - Referee Calcolo LHC
CNAF • Punto focale di tutte le attività INFNGRID • Manpower finanziato sui progetti • Tier1 per esperimenti LHC • Funzionante e pienamente utilizzato • Manpower INFN fortemente carente • Sia per la gestione, sia per l’upgrade infrastrutturale • Indicazione dei referee: • Concentrarsi sulle attività di Core GRID necessarie per il calcolo degli esperimenti LHC • Questione delicata per i contributi approvati dall’INFN ai progetti internazionali • Piano di sviluppo ancora in discussione. Elementi da definire: • Necessità degli esperimenti nel 2007-2008 • Espandibilità del CNAF nel 2007 • Le infrastrutture esistenti mostrano forti limiti sia per il condizionamento sia per la distribuzione elettrica • Interventi urgenti previsti in parallelo all’upgrade infrastrutturale completo (che non si concluderà prima di primavera 2008) • Le risorse pledged a WLCG per il 2007 non sembrano raggiungibili F.Forti - Referee Calcolo LHC
Risorse fornite vs. pledged F.Forti - Referee Calcolo LHC
Suddivisione risorse CNAF • Comitato di gestione • Gruppo deputato a definire la suddivisione delle risorse e le altre scelte operative del centro • Da utilizzare in modo sistematico e continuativo • Nuovo coordinatore delle richieste degli esperimenti è Umberto Marconi (grazie Paolo) • Richieste da parte di esperimenti di CSN2 • Argo, Virgo, Pamela, Opera, Magic • Soprattutto spazio disco: critico perchè gli esperimenti sono in presa dati • Da privilegiare l’acquisto di disco, che consuma anche meno potenza... • Incontro con la commissione II domani • Tutti gli esperimenti che calcolano al CNAF dovrebbero essere referati nello stesso gruppo • Attualmente Babar e CDF sono a parte F.Forti - Referee Calcolo LHC
Tier2 • I Tier2 sono finanziati per il 2007 esclusivamente con fondi SJ. • Gli esperimenti stanno preparando un piano dettagliato di attività per il 2007 in modo da definire gli sblocchi di SJ • Pronto verso fine anno • Esaminati in dettaglio i progetti di Roma (Torino) e Pisa • Relazione di referaggio di CCR (26/10/06): R. Gomezel, M. Morandin, L. Pellegrino, R. Ricci, R. Stroili • In generale sono stati fatti notevoli progressi • L’efficienza di utilizzo e la collaborazione tra TIER2 cresciuti notevolmente F.Forti - Referee Calcolo LHC
T2-Roma • Il progetto ha raggiunto un livello globalmente adeguato di approfondimento delle questioni tecniche, tale da poter permettere il passaggio alla fase esecutiva. • Il gruppo di Roma ha sfruttato competenze locali esistenti, si è appoggiato al servizio di LNL e inoltre si servirà per il progetto esecutivo di una ditta specializzata già individuata. • La questione tecnica risultata di più difficile soluzione, ovvero la collocazione delle macchine condensanti esterne, sembra ora risolta. • Le questioni tecniche critiche sono state affrontate e le opzioni tecniche presentate non presentano particolari rischi. • Si nota la mancanza di un documento preliminare di progettazione per gli impianti elettrici che invece è stato reso disponibile per la parte di condizionamento. • Da fare: • progetto di massima del sistema di gestione dei guasti critici, • integrazione impiantistica del sistema antiincendio della sala • progetto aggiornato della rete locale F.Forti - Referee Calcolo LHC
T2-Pisa • Il progetto ha subito modifiche importanti rispetto alla sua prima formulazione. • Nella sua versione attuale, con il trasferimento di parte delle macchine ad una nuova sala, si sono ottenuti considerevoli semplificazioni e risparmi. • I principali elementi del progetto sono stati definiti e le soluzioni tecniche proposte sembrano in generale adeguate. • I documenti forniti esaminano in dettaglio gli aspetti critici, ma, per servire come base per una progettazione esecutiva, andrebbero ulteriormente integrati con le informazioni che ora risultano mancanti. • Il gruppo si appoggia ai tecnologi disponibili in Sezione e non prevede di coinvolgere professionisti esterni per stilare il progetto esecutivo. • Da fare: • Condizionamento: • Documento dettagliato per l’affidamento della progettazione • Ottimizzazione e ridondanza dell’impianto • Analisi guasti critici • Impianti elettrici • tabella dei carichi elettrici includendo tutte le utenze (SNS; Dipartimento) • Riconsiderae i margini di potenza, che sembrano molto stretti • Analisi dell’affidabilità del sistema • Riconsiderare la scelta di non utilizzare un UPS per una parte delle macchine F.Forti - Referee Calcolo LHC
Piano dei TIER2 • Il piano dettagliato dei TIER2 verrà esaminato a gennaio • Si vogliono finanziare le attività, non le tabelle • Per i TIER2 ancora SJ (Milano e Pisa) • Come detto a settembre le condizioni tecniche sono ragionevolmente soddisfatte • (a volte funzionano meglio dei Tier2 approvati) • Le comunità di riferimento sono attive • Rimane la questione generale dell’effettiva necessità di calcolo di LHC • Necessario finanziare anche i T2 SJ ad un livello sufficiente a sopravvivere F.Forti - Referee Calcolo LHC
Conclusioni • La matassa del calcolo LHC è certamente intricata • La GRID ha dimostrato che potenzialmente può risolvere il problema, anche se la performance non è ancora sufficiente • Rispetto al passato, maggiore enfasi su reliability e availability piuttosto che su nuove caratteristiche • L’INFN è piazzato centralmente in questa attività e contribuisce moltissimo • E’ essenziale risolvere al più presto le difficoltà infrastrutturali del CNAF per farlo operare a pieno ritmo • E’ necessario focalizzare tutte le forze per la realizzazione del calcolo LHC, anche se questo può limitare altre attività interessanti • Le sezioni rappresentano un serbatoio vitale di idee e persone per far avanzare il programma e devono essere pienamente coinvolte F.Forti - Referee Calcolo LHC