Distributed Computing in IHEP

Slides:



Advertisements
Similar presentations
Status of BESIII Distributed Computing BESIII Workshop, Mar 2015 Xianghu Zhao On Behalf of the BESIII Distributed Computing Group.
Advertisements

1 Bridging Clouds with CernVM: ATLAS/PanDA example Wenjing Wu
LHC Experiment Dashboard Main areas covered by the Experiment Dashboard: Data processing monitoring (job monitoring) Data transfer monitoring Site/service.
Distributed Computing for CEPC YAN Tian On Behalf of Distributed Computing Group, CC, IHEP for 4 th CEPC Collaboration Meeting, Sep ,
BESIII physical offline data analysis on virtualization platform Qiulan Huang Computing Center, IHEP,CAS CHEP 2015.
BESIII distributed computing and VMDIRAC
YAN, Tian On behalf of distributed computing group Institute of High Energy Physics (IHEP), CAS, China CHEP-2015, Apr th, OIST, Okinawa.
BESIII Production with Distributed Computing Xiaomei Zhang, Tian Yan, Xianghu Zhao Institute of High Energy Physics, Chinese Academy of Sciences, Beijing.
NOVA Networked Object-based EnVironment for Analysis P. Nevski, A. Vaniachine, T. Wenaus NOVA is a project to develop distributed object oriented physics.
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.
Distributed Computing for CEPC YAN Tian On Behalf of Distributed Computing Group, CC, IHEP for 4 th CEPC Collaboration Meeting, Sep , 2014 Draft.
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.
INFSO-RI Enabling Grids for E-sciencE ARDA Experiment Dashboard Ricardo Rocha (ARDA – CERN) on behalf of the Dashboard Team.
DIRAC Pilot Jobs A. Casajus, R. Graciani, A. Tsaregorodtsev for the LHCb DIRAC team Pilot Framework and the DIRAC WMS DIRAC Workload Management System.
Status of BESIII Distributed Computing BESIII Workshop, Sep 2014 Xianghu Zhao On Behalf of the BESIII Distributed Computing Group.
DIRAC Project A.Tsaregorodtsev (CPPM) on behalf of the LHCb DIRAC team A Community Grid Solution The DIRAC (Distributed Infrastructure with Remote Agent.
1 A Scalable Distributed Data Management System for ATLAS David Cameron CERN CHEP 2006 Mumbai, India.
Distributed Physics Analysis Past, Present, and Future Kaushik De University of Texas at Arlington (ATLAS & D0 Collaborations) ICHEP’06, Moscow July 29,
The GridPP DIRAC project DIRAC for non-LHC communities.
Meeting with University of Malta| CERN, May 18, 2015 | Predrag Buncic ALICE Computing in Run 2+ P. Buncic 1.
Virtual Cluster Computing in IHEPCloud Haibo Li, Yaodong Cheng, Jingyan Shi, Tao Cui Computer Center, IHEP HEPIX Spring 2016.
A Data Handling System for Modern and Future Fermilab Experiments Robert Illingworth Fermilab Scientific Computing Division.
Breaking the frontiers of the Grid R. Graciani EGI TF 2012.
StoRM + Lustre Proposal YAN Tian On behalf of Distributed Computing Group
Multi-community e-Science service connecting grids & clouds R. Graciani 1, V. Méndez 2, T. Fifield 3, A. Tsaregordtsev 4 1 University of Barcelona 2 University.
Status of BESIII Distributed Computing BESIII Collaboration Meeting, Nov 2014 Xiaomei Zhang On Behalf of the BESIII Distributed Computing Group.
Vendredi 27 avril 2007 Management of ATLAS CC-IN2P3 Specificities, issues and advice.
The CMS Beijing Tier 2: Status and Application Xiaomei Zhang CMS IHEP Group Meeting December 28, 2007.
Scientific Data Processing Portal and Heterogeneous Computing Resources at NRC “Kurchatov Institute” V. Aulov, D. Drizhuk, A. Klimentov, R. Mashinistov,
Distributed computing and Cloud Shandong University (JiNan) BESIII CGEM Cloud computing Summer School July 18~ July 23, 2016 Xiaomei Zhang 1.
DIRAC: Workload Management System Garonne Vincent, Tsaregorodtsev Andrei, Centre de Physique des Particules de Marseille Stockes-rees Ian, University of.
CEPC software & computing study group report
CernVM-FS vs Dataset Sharing
Dynamic Extension of the INFN Tier-1 on external resources
WLCG IPv6 deployment strategy
Review of the WLCG experiments compute plans
DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING CLOUD COMPUTING
Grid Computing: Running your Jobs around the World
Status of BESIII Distributed Computing
Status of BESIII Computing
The advances in IHEP Cloud facility
Xiaomei Zhang CMS IHEP Group Meeting December
AWS Integration in Distributed Computing
Elastic Computing Resource Management Based on HTCondor
Computing models, facilities, distributed computing
Overview of the Belle II computing
Report of Dubna discussion
ATLAS Cloud Operations
ALICE Monitoring
Yaodong CHENG Computing Center, IHEP, CAS 2016 Fall HEPiX Workshop
DIRAC services.
Joseph JaJa, Mike Smorul, and Sangchul Song
LHCb Computing Model and Data Handling Angelo Carbone 5° workshop italiano sulla fisica p-p ad LHC 31st January 2008.
Readiness of ATLAS Computing - A personal view
Computing at CEPC Xiaomei Zhang Xianghu Zhao
Job workflow Pre production operations:
Discussions on group meeting
WLCG Collaboration Workshop;
VMDIRAC status Vanessa HAMAR CC-IN2P3.
The Scheduling Strategy and Experience of IHEP HTCondor Cluster
Xiaomei Zhang On behalf of CEPC software & computing group Nov 6, 2017
Haiyan Meng and Douglas Thain
ExaO: Software Defined Data Distribution for Exascale Sciences
AIMS Equipment & Automation monitoring solution
Status and prospects of computing at IHEP
Status and plans for bookkeeping system and production tools
Exploit the massive Volunteer Computing resource for HEP computation
Exploring Multi-Core on
The LHCb Computing Data Challenge DC06
Presentation transcript:

Distributed Computing in IHEP Xiaomei Zhang On behalf of IHEP distributed computing group 2017 Spring Hepix , Budpest

Motivation Increasing experiments and data volume in IHEP Put pressure on one single data center Extra resources needed as a supplement Possible resource contributing from wide international cooperation in experiments Various heterogeneous opportunistic resources existing Distributed computing is the way to integrate as a whole Distributed and heterogeneous resources

BESIII (Beijing Spectrometer III at BEPCII) A bit history Distributed Computing (DC) in IHEP was first built in 2012 Meet peek need of BESIII with ~3PB/5year Put into production in 2014 Resources at beginning: 80% Batch, 20% Grid Integrate cloud in 2015 Resources now: 10%Grid, 65%Batch, 25%Clouds BESIII (Beijing Spectrometer III at BEPCII)

CEPC Collider Ring(50Km) A bit history In 2015, evolved into a general platform for multi experiments More new experiments are coming in IHEP JUNO, LHAASO, CEPC…… More than one experiments express interests on using or evaluating Save manpower and simplify management BTC IP1 IP2 e+ e- e+ e- Linac (240m) LTB CEPC Collider Ring(50Km) Booster(50Km) LHAASO Large High Altitude Air Shower Observatory CEPC JUNO (Jiangmen Underground Neutrino Observatory)

Computing model IHEP as central site Remote sites Data flow Raw data processing, bulk reconstruction, analysis Remote sites MC production, analysis Sites without SE only for MC, job output directly to remote SEs Data flow Central storage in IHEP IHEP -> Sites, DST for analysis Sites -> IHEP, MC data for backup Comparing LCG, size and man power is small. We keep everything simple to make sites and central management as easy as possible with small working group Simple, easy with small group Most of sites without grid experience

Resource Site: 15 from USA, Italy, Russia, Turkey, Taiwan, China Universities(8) Network: 10Gb/s to USA and Europe, 10 Gb/s to TaiWan and Mainland Joining LHCONE is in plan to future improve network Bottleneck in end-to-end, PerfSonar monitoring in plan Resource: ~3000 CPU cores, ~500TB storage Job input and output directly from remote SE IHEP 10Gb/s 10Gb/s 10Gb/s

DIRAC-based WMS DIRAC (Distributed Infrastructure with Remote Agent Control) Middle layer between jobs and heterogeneous resources GANGA and JSUB for massive job submission and management JSUB, newly developed for flexible workflow and general purpose CVMFS (CERN VM FileSystem) deploy experiment software to remote sites

CVMFS set-up in IHEP In 2013, the repo boss.cern.ch for BESIII created at CERN In 2015, IHEP CVMFS Stratum0(S0) created Support other experiments including CEPC, JUNO, LHAASO…… 3 repositories, ~600GB In 2017 New IHEP S0 with HA created for both DC and local batch In 2017, IHEP CVMFS Stratum1(S1) created, both serve IHEP S0 and CERN (RAL) S0 Speed LHC and non-LHC software access in Asia Plan to have S1 outside Asia to speed up access of IHEP software

JSUB (Job submission tool) Lightweight and general framework developed to take care of life cycle of tasks (a bunch of jobs) Extensible arch with plug-ins make easy for experiments to create own Modular designs allow job workflow to be customizable “Step” and “Module” can be reused

FroNtier/squid for offline DB access Static SQLite DB on CVMFS and Mirror DB are in use for DC FroNtier/Squid based on Cache tech is considered Provide real-time and more stable database access in DC env On-going works Allow Frontier to accept MySQL to be backend Bind Frontier with experiment software to allow transparent access

Multi-VO control in DIRAC VOMS set up to group experiments cepc, juno, bes Resource scheduling and control Users grouped with VOs Jobs can get VO info from owners Pilots separated with VOs Pull only same VO jobs Resources tagged with VOs Name the owners Scheduler can do the matching and priority control based on VO and roles Different colors for Different VOs

Metadata and File catalogue Built based on DFC (DIRAC File Catalogue) Directory-like usage, tightly coupled with DIRAC WMS Combine Replica, Metadata and Dataset Catalogue Performance similar to LFC+AMGA, more convenient Permission control through DIRAC user management Allow separated usage and control for each experiment Current data registered Size: ~300TB

Central SE with StoRM Start with dCache, and Switch to StoRM in 2015 Simply data exchange between local and grid One set-up to allow multi-experiments Frontend(SRM, HTTP, xrootd) + Backend(Lustre) Frontend, permission control and role matching with VOMS Backend, local file systems with experiment data Performance fine with current load Version: StoRM1. 11. 11 Capacity: ~2.5PB Guide to help small sites to have SE with StoRM Lustre Capacity(TB) Model owner /gridfs 66 RW public /bes3fs 1100 RO bes /juofs 502 juno /cefs 794 cepc

Massive data transfer system Developed as a DIRAC service Share data among sites, Transfer between site SEs Dataset supported, multi-streams design for high speed Permission control through DIRAC user management Each year, ~100TB data transfers between sites

Elastic cloud integration Implement cloud integration in an elastic way Extension VMDIRAC with VM scheduler was introduced VM booted with “pilot” to pull jobs from TaskQueue VM contextualization done with Cloud-init Cloud types supported OpenStack, OpenNebula, AWS Interface used: libcloud, rOCCI, boto Not easy to find a general layer to meet all requirements

Elastic cloud integration Become important part of IHEP DC 6 cloud sites are available from Italy, China, Russia, More than 700K jobs have been processed in the past 2 years ~ 5% failure related with VM performance issue VM got stuck, not easy to track down

Commercial cloud exploration In 2015.6, AWS cloud integrated in elastic way as others Trial done with the support of Amazon AWS China region BES image created and upload to AWS Connect with AWS API in VMDIRAC elastic scheduling Tests done and Price evaluated 400K BES rhopi events simulated with 100% success rate simu+reco+ana with few output, mainly focus on evaluating CPU price Proper CPU type need to be chosen for best performance and price Price is still a bit high (~10) comparing to self-maintenance batch Use Amazon market price in China

Action-based Site Monitoring Motivation Improve site stability Ease the life of admins A global site status view Components Information Collection and Display Decision and actions Both Active and Passive info collected Policies defined for automatic actions taken in case of problems Sending warning messages, ban sites

Production status Total Jobs are 728K in 2016, 665K in 2015, 340K in 2014 Max running jobs can reach 2K (First season in 2015) Data exchange about 300TB directly from jobs each year

Multi-core supports Multi-process/thread applications are booming in HEP Best exploitation of multicore CPU architectures Decrease memory usage per core Multi-core(Mcore) supports are considered Current: One-core pilot pull one-core job First way: Mcore pilots added to pull Mcore jobs M-core pilots pull M-core jobs Easy to implement Pilot “starving” when matching with mixture of n-core job (n=1, 2…)

Multi-core supports Second way: Standard-size pilots with dynamic patitionable slots Standard-size: whole-node, 4-node, 8-node…. Pilots pull n-core jobs (n=1, 2…..) until internal slots used up More complicated to implement Schedule efficiency is more complicated than single-core How many pliots needed? Best match-making of jobs to pilots? Pilot Job pool

Future Plan HPC federation with DIRAC started to build a “grid” of HPC computing resources HPC resource becomes more and more important in high energy physics data processing Many HPC computing centers are being built up in recent years among HEP data centers Scaling and related performance study are in considerations to meet possible challenge of large experiments Data Federation with Cache considered Speed up data access in sites Free small sites from maintenance of storage

Summary Mature techs used as much as possible to keep easy and simple for sites and central manager Developments done to meet specific needs of our experiments Small scale and work fine with current resource and load Keep up with advanced techs to meet future challenge and more requirements from experiments

Thank you!