Maria Girone, CERN CMS Experiment Status, Run II Plans, & Federated Requirements Maria Girone, CERN XrootD Workshop, January 27, 2015.

Slides:



Advertisements
Similar presentations
Introduction to CMS computing CMS for summer students 7/7/09 Oliver Gutsche, Fermilab.
Advertisements

Distributed Xrootd Derek Weitzel & Brian Bockelman.
T1 at LBL/NERSC/OAK RIDGE General principles. RAW data flow T0 disk buffer DAQ & HLT CERN Tape AliEn FC Raw data Condition & Calibration & data DB disk.
Ian M. Fisk Fermilab February 23, Global Schedule External Items ➨ gLite 3.0 is released for pre-production in mid-April ➨ gLite 3.0 is rolled onto.
Large scale data flow in local and GRID environment V.Kolosov, I.Korolko, S.Makarychev ITEP Moscow.
Ian Fisk and Maria Girone Improvements in the CMS Computing System from Run2 CHEP 2015 Ian Fisk and Maria Girone For CMS Collaboration.
Status of CMS Matthew Nguyen Recontres LCG-France December 1 st, 2014 *Mostly based on information from CMS Offline & Computing Week November 3-7.
Zhiling Chen (IPP-ETHZ) Doktorandenseminar June, 4 th, 2009.
Test Of Distributed Data Quality Monitoring Of CMS Tracker Dataset H->ZZ->2e2mu with PileUp - 10,000 events ( ~ 50,000 hits for events) The monitoring.
Grid Data Management A network of computers forming prototype grids currently operate across Britain and the rest of the world, working on the data challenges.
Computing and LHCb Raja Nandakumar. The LHCb experiment  Universe is made of matter  Still not clear why  Andrei Sakharov’s theory of cp-violation.
Claudio Grandi INFN Bologna CMS Operations Update Ian Fisk, Claudio Grandi 1.
CHEP – Mumbai, February 2006 The LCG Service Challenges Focus on SC3 Re-run; Outlook for 2006 Jamie Shiers, LCG Service Manager.
Computing Infrastructure Status. LHCb Computing Status LHCb LHCC mini-review, February The LHCb Computing Model: a reminder m Simulation is using.
INTRODUCTION The GRID Data Center at INFN Pisa hosts a big Tier2 for the CMS experiment, together with local usage from other HEP related/not related activities.
Take on messages from Lecture 1 LHC Computing has been well sized to handle the production and analysis needs of LHC (very high data rates and throughputs)
Your university or experiment logo here Caitriana Nicholson University of Glasgow Dynamic Data Replication in LCG 2008.
1. Maria Girone, CERN  Q WLCG Resource Utilization  Commissioning the HLT for data reprocessing and MC production  Preparing for Run II  Data.
ATLAS in LHCC report from ATLAS –ATLAS Distributed Computing has been working at large scale Thanks to great efforts from shifters.
Experiment Support CERN IT Department CH-1211 Geneva 23 Switzerland t DBES Successful Common Projects: Structures and Processes WLCG Management.
November SC06 Tampa F.Fanzago CRAB a user-friendly tool for CMS distributed analysis Federica Fanzago INFN-PADOVA for CRAB team.
Tier-2  Data Analysis  MC simulation  Import data from Tier-1 and export MC data CMS GRID COMPUTING AT THE SPANISH TIER-1 AND TIER-2 SITES P. Garcia-Abia.
CERN IT Department CH-1211 Genève 23 Switzerland t Frédéric Hemmer IT Department Head - CERN 23 rd August 2010 Status of LHC Computing from.
Performance Tests of DPM Sites for CMS AAA Federica Fanzago on behalf of the AAA team.
Claudio Grandi INFN Bologna CMS Computing Model Evolution Claudio Grandi INFN Bologna On behalf of the CMS Collaboration.
6/23/2005 R. GARDNER OSG Baseline Services 1 OSG Baseline Services In my talk I’d like to discuss two questions:  What capabilities are we aiming for.
High Availability in DB2 Nishant Sinha
SLACFederated Storage Workshop Summary For pre-GDB (Data Access) Meeting 5/13/14 Andrew Hanushevsky SLAC National Accelerator Laboratory.
The CMS CERN Analysis Facility (CAF) Peter Kreuzer (RWTH Aachen) - Stephen Gowdy (CERN), Jose Afonso Sanches (UERJ Brazil) on behalf.
US-CMS T2 Centers US-CMS Tier 2 Report Patricia McBride Fermilab GDB Meeting August 31, 2007 Triumf - Vancouver.
LHCbComputing LHCC status report. Operations June 2014 to September m Running jobs by activity o Montecarlo simulation continues as main activity.
David Stickland CMS Core Software and Computing
Predrag Buncic ALICE Status Report LHCC Referee Meeting CERN
ATLAS Distributed Computing perspectives for Run-2 Simone Campana CERN-IT/SDC on behalf of ADC.
Efi.uchicago.edu ci.uchicago.edu Ramping up FAX and WAN direct access Rob Gardner on behalf of the atlas-adc-federated-xrootd working group Computation.
1 A Scalable Distributed Data Management System for ATLAS David Cameron CERN CHEP 2006 Mumbai, India.
Ian Bird Overview Board; CERN, 8 th March 2013 March 6, 2013
LHCbComputing Computing for the LHCb Upgrade. 2 LHCb Upgrade: goal and timescale m LHCb upgrade will be operational after LS2 (~2020) m Increase significantly.
Latest Improvements in the PROOF system Bleeding Edge Physics with Bleeding Edge Computing Fons Rademakers, Gerri Ganis, Jan Iwaszkiewicz CERN.
Distributed Physics Analysis Past, Present, and Future Kaushik De University of Texas at Arlington (ATLAS & D0 Collaborations) ICHEP’06, Moscow July 29,
GDB, 07/06/06 CMS Centre Roles à CMS data hierarchy: n RAW (1.5/2MB) -> RECO (0.2/0.4MB) -> AOD (50kB)-> TAG à Tier-0 role: n First-pass.
CMS: T1 Disk/Tape separation Nicolò Magini, CERN IT/SDC Oliver Gutsche, FNAL November 11 th 2013.
Markus Frank (CERN) & Albert Puig (UB).  An opportunity (Motivation)  Adopted approach  Implementation specifics  Status  Conclusions 2.
ALICE Grid operations +some specific for T2s US-ALICE Grid operations review 7 March 2014 Latchezar Betev 1.
Meeting with University of Malta| CERN, May 18, 2015 | Predrag Buncic ALICE Computing in Run 2+ P. Buncic 1.
Monitoring the Readiness and Utilization of the Distributed CMS Computing Facilities XVIII International Conference on Computing in High Energy and Nuclear.
Maria Girone, CERN  CMS in a High-Latency Environment  CMSSW I/O Optimizations for High Latency  CPU efficiency in a real world environment  HLT 
1 June 11/Ian Fisk CMS Model and the Network Ian Fisk.
Experiment Support CERN IT Department CH-1211 Geneva 23 Switzerland t DBES The Common Solutions Strategy of the Experiment Support group.
Alessandro De Salvo CCR Workshop, ATLAS Computing Alessandro De Salvo CCR Workshop,
Any Data, Anytime, Anywhere Dan Bradley representing the AAA Team At OSG All Hands Meeting March 2013, Indianapolis.
Joe Foster 1 Two questions about datasets: –How do you find datasets with the processes, cuts, conditions you need for your analysis? –How do.
Apr. 25, 2002Why DØRAC? DØRAC FTFM, Jae Yu 1 What do we want DØ Regional Analysis Centers (DØRAC) do? Why do we need a DØRAC? What do we want a DØRAC do?
Efi.uchicago.edu ci.uchicago.edu Sharing Network Resources Ilija Vukotic Computation and Enrico Fermi Institutes University of Chicago Federated Storage.
Computing infrastructures for the LHC: current status and challenges of the High Luminosity LHC future Worldwide LHC Computing Grid (WLCG): Distributed.
LHCb LHCb GRID SOLUTION TM Recent and planned changes to the LHCb computing model Marco Cattaneo, Philippe Charpentier, Peter Clarke, Stefan Roiser.
DATA ACCESS and DATA MANAGEMENT CHALLENGES in CMS.
ScotGRID is the Scottish prototype Tier 2 Centre for LHCb and ATLAS computing resources. It uses a novel distributed architecture and cutting-edge technology,
ATLAS – statements of interest (1) A degree of hierarchy between the different computing facilities, with distinct roles at each level –Event filter Online.
Efi.uchicago.edu ci.uchicago.edu FAX splinter session Rob Gardner Computation and Enrico Fermi Institutes University of Chicago ATLAS Tier 1 / Tier 2 /
1. Maria Girone, CERN  Highlights of Run I  Challenges of 2015  Mitigation for Run II  Improvements in Data Management, Organized Processing and Analysis.
Maria Girone, CERN CMS Status Report Maria Girone, CERN David Lange, LLNL.
SuperB and its computing requirements
for the Offline and Computing groups
LHCb Computing Model and Data Handling Angelo Carbone 5° workshop italiano sulla fisica p-p ad LHC 31st January 2008.
Dagmar Adamova (NPI AS CR Prague/Rez) and Maarten Litmaath (CERN)
Evolution of the distributed computing model The case of CMS
Thoughts on Computing Upgrade Activities
Ákos Frohner EGEE'08 September 2008
ALICE Computing Model in Run3
Presentation transcript:

Maria Girone, CERN CMS Experiment Status, Run II Plans, & Federated Requirements Maria Girone, CERN XrootD Workshop, January 27, 2015

Maria Girone, CERN  Computing Status and Preparation for Run II  CMS Computing Improvements during Ls1  Data Management  Data Placement  Data Access  Federation Progress and Requirements  Validation and Commissioning during the Computing Software and Analysis Challenge (CSA14) and current state of the federation  Outlook 2

Maria Girone, CERN  In Run II, CMS expects to promptly reconstruct 1000Hz of data (a factor 2 more) at an average of twice the reconstruction time (another factor of 2 more)  Sample of events to collect, promptly reconstruct and eventually reprocess is 3B events in 2015  CMS expects to generate 1.3 times simulated events for every event collected (4B GEN-SIM by end of 2015)  Budget for processing is ~1.5 times events collected  Analysis facilities were specified to primarily provide resources to analyze Run II data only  The resource request assumes that Run I data analysis will ramp down with the start of the run 3

Maria Girone, CERN  Organized processing for data and simulation reconstruction and reprocessing capacity will nearly double wrt 2012, while processing needs increase by more than a factor of 4  We need to do more with less, use resources more efficiently and in a more flexible way  Disk resources double at Tier-0, but increase only 15% at Tier- 1s and Tier-2s, despite the doubling of the trigger rate and the increase in event size  The expectation is that we must use the storage much more effectively  The data federation is a crucial element in this 4

Maria Girone, CERN  CERN  12k cores for Tier-0 and additional 15k when the HLT farm is available  15PB of disk and 31PB of tape  Network between Gb/s to Tier1s  Tier-1  7 facilities primarily at national labs or large computing centers  ~40k cores 27PB of disk, and 74PB of tape  Network between 1Gb/s – 100Gb/s to Tier2s  Tier-2  50 facilities primarily at university centers  80k cores and 31PB of disk CERN T0/HLT Tier-1 Tier- 2 5

Maria Girone, CERN  In LS1 we have made 2 significant changes  Logically the disk and tape storage systems have been split  A file written to disk on a Tier-1 site needs to be explicitly subscribed to a tape endpoint  A data federation has been deployed across CERN, all Tier-1s, and 90% of the Tier-2 disk space  Hierarchical redirectors have been installed 6  Boundaries between sites have been removed  The difference between capability of all the Tiers has been reduced  Improved networks are key to this

Maria Girone, CERN Concretely these two changes have a big impact on the how resources could be used (additional capabilities in dash-lines) HLT Tier-1 Tier-0 Tier-2 GEN-SIM MC RECO DATA RECO ANALYSIS 7

Maria Girone, CERN  An addition for Run II is the use of the High Level Trigger (HLT) farm for offline processing  It is a large computing resource (15k cores) that is similar in size to the Tier-0 in terms of number of cores, but we cannot reach this scale until March  Successfully interfaced using cloud computing tools. It is similar to the Tier-0 AI  In 2014 the network link P5 to the computing center was upgraded from 20 to 60Gb/s  Far larger than needed for data taking but necessary to access the storage in the computing center for simulation reconstruction  Will be upgraded to 120Gb/s before the 2015 run starts  Production workflows have been commissioned including the HI reprocessing, Gen-Sim, and Simulation reconstruction  All access to data is through the data federation and primarily served from CERN 40Gbit/s sustained 8

Maria Girone, CERN  In addition to data federation, we have improved our traditional data placement and access  The use of samples is continuously monitored through the data popularity  A new miniAOD format has been added for 80% of the analysis in 10% of the size  Samples will be replicated dynamically if the load is high and replicas removed if they are not accessed for a period of time  Data and MC samples are distributed to Tier-2 (Tier-1) sites automatically  Number of replicas depends on the popularity of the datasets  Samples are only deleted when there is new data to replicate, disks are kept full  We are working to improve and automate the monitoring of the disk usage  This is needed in part because the storage is inefficiently used  Even at a 6M window, 27% of the disk space for AODSIM is used by un-accessed samples (10% of the total space)  The zero bin includes un-accessed replicas and datasets that have only one copy on disk 9

Maria Girone, CERN  Access to data is CMS is done in 2 ways  At a local site with an interactive application by specifying the Logical file name (LFN), which is resolved into a file to open  This is done by users and debuggers  Or remote submission, where a job is specified by dataset (a large group of LFNs) and a large number of jobs is launched to grid enabled resources.  Distributed analysis and production  There have been significant improvements in both with the commissioning of a data federation 10

Maria Girone, CERN  System Requirements  Provide remote access to CERN, Tier-1, and nearly all Tier-2 disk storage through XrootD  More than 90% of all current CMS data should be accessible  Sufficient IO capacity to provide 20% of the total access (~100TB/day)  Enable to system of hierarchical redirectors to maintain access within geographic regions when possible  Functionality Requirements  Interactive access  User opens a file from anywhere. Performance and stability are secondary to convenience  Failover protection  A cluster local storage fails to deliver a file and the federation serves as a backup. These jobs were going to die and the federation can save some  Overflow  A site is busy and jobs are intentionally redirected to a site nearby. Analysis Operations chooses close reliable sites  A user intentionally directs jobs to any site and accesses samples over the wide area at her/his own risk  Production  Production jobs share workflows across sites and the federation is used to serve the data. In order to have good operational efficiency the federation has to have similar performance and reliability to local storage 11

Maria Girone, CERN  Any Data, Anytime, Anywhere (AAA) has been a primary focus area in 2014  CERN, all Tier-1s, most of the Tier-2 sites serve data in the federation  Nearly all sites are configured to use the federation to access samples, if they aren’t available locally  Optimization of the IO has been an ongoing activity for several years, which has paid off in high CPU efficiency over the wide area  Big push in 2014 to commission sites to measure IO and file open rates and to understand the sustainable load and to deploy and use advanced monitoring  Target for Run II is that 20% of all access is through the federation 12

Maria Girone, CERN  We validated small scale use of non-local data access in CSA14  Fall-back when CRAB3 jobs don't find input data locally and in "ignore locality" mode  Very good feedback by users  After CSA14 scale tests were performed in Europe and the US  20% of jobs were able to access data over the wide area (60k files/day, O(100TB)/day)  Tests showed that the scale could be reached, but that the job success rates were sensitive to the health of all the sites 13

Maria Girone, CERN  The addition of the production workflow puts additional constraints on the required reliability of the Federation  Validated sites are in the production federation, sites being commissioned are in an independent federation and only when a sample cannot be found in production are they used Redirector Site Redirector Site Production Transitional 14

Maria Girone, CERN  In Run I most of the low latency commissioning and analysis activities had to be done on the CAF (CERN Analysis Facility)  The express data is processed and delivered in an hour, but that doesn’t allow time for merging so the data could not be efficiently replicated to remote sites  Access to the storage and CPU on the CAF was limited to about 10% of the collaboration  In 2015 CERN will have 12PB of data serving local users and the federation  The express data, in addition to all samples, will be accessible to everyone as soon as it’s produced  The federation equalizes access to samples 15

Maria Girone, CERN  Currently all sites in CMS have the failover protection enabled  A small site configuration change protects against site storage failures  It’s only a few percent of jobs, but good to have protection  Overflow is enabled for US sites; we are preparing a European deployment  A matrix of well connected sites is created and jobs are directed to alternative locations when the queuing time becomes too large  Normal operations is ~10% of the jobs, and this improves the average waiting time 16

Maria Girone, CERN The use of Data Federation in production is still under commissioning Proof of Concept Last January during a storage update, FNAL accidently pointed to an obsolete DB and production traffic failed over to Xrootd – Data was read from CNAF disk at 400MB/s for 2 days – Production jobs continued to succeed at the same rate but the network monitors noticed traffic with >3Gb/s on the OPN FNALCNAF This was a good scale test for AAA for shared production Benefited from WN on OPN (case of FNAL) 17

Maria Girone, CERN  After significant software development, CMS now has a production ready multi- core framework  More than 99% of the running core can run in parallel. This allows us to take better advantage of the growth in cores per system, decrease the memory footprint, reduced the number of running processes to track, improved merging, and simplified the IO paths  Big memory savings: 0.35 GB per additional thread instead of 1.8GB/job An interesting side effect of this is that the IO of the application increases by a factor of 4-8 depending on the number of cores used Instead of many small IO processes (50-100kB/s for Reco) we have fewer processes with higher IO (200kB/s-1MB/s) For DIGI 3-5MB/s becomes 12-40MB/s 18

Maria Girone, CERN  One of the challenges of opportunistic resources is access to data  Batch queues are dynamic and access is given. Storage usage is more static and resources are not given temporarily, because people are much more willing to kill jobs than to delete samples  Setting up a storage element an a PhEDEx instance is labor intensive and doesn’t lend itself to dynamic deployment  Up to now this has limited to kind of processing we did opportunistically to activities that didn’t require much input samples  The data federation allows data to be streamed to opportunistic CPUs and opens up a class of activities on opportunistic computing 19

Maria Girone, CERN  We believe that access to the samples will be dramatically improved in Run II  Use of data federation and dynamic data management should ensure that users have access to everything from wherever they choose to work  Run II will be a challenge and improvements in data access were necessary to make the best use of the resources  We will have less computing and storage per event collected and simulated than we did in Run I  We need to improve efficiency  Data Federation improves data access for everyone and improves the flexibility for using resources for organized processing  Feedback from users has been extremely positive 20