tiers_grid2012

Size: px

Start display at page:

Download "tiers_grid2012"

응수 문
5 years ago
Views:

1 Tiers and GRID computing ( )

today only possible due to extraordinary performance of accelerators - experiments -

2 The higgs, the history, and the grid 28 years since the idea of the LHC thousands of people worldwide thousands of computers worldwide Global Effort Global Success Results today only possible due to extraordinary performance of accelerators - experiments - Grid computing July 4, 2012: a final comment from Rolf Heuer, the director general of CERN 2

Dec 25, 1990: the first successful communication between a Hypertext Transfer Protocol

3 CERN Physics Particles, energy and matter to understand more about how our universe works Education Find new ideas on bringing modern physics into the classroom Computing Dec 25, 1990: the first successful communication between a Hypertext Transfer Protocol (HTTP) and server via the Internet by Tim Berners-Lee, the inventor of the World Wide Web 3

CERN in our universe Largest and biggest New particles in high-energy collisions LHC smashes groups of protons together at close to the speed of light: 40 million times per

4 CERN in our universe Largest and biggest New particles in high-energy collisions LHC smashes groups of protons together at close to the speed of light: 40 million times per second P5 where you visited for CMS detector Fastest Hottest spots = 10 5 Coldest and emptiest space (1.9 K) Most powerful computer system 10 5 dual layer DVDs/year = PetaBytes/yr

computing network called the Grid Data Analysis

5 Computing for Physics Research Data Technologies Data storing and management distributed, parallel and cloud computing network called the Grid Data Analysis Algorithm and tools simulation, reconstruction and visualization

Computing system Grids (a super virtual computer) A supercomputer

interface, such as Ethernet Geographically distributed computing

is a subsystem that transfers data between components Europe

6 Computing system Grids (a super virtual computer) A supercomputer Have computers connected to a network by a conventional network interface, such as Ethernet Geographically distributed computing Has many processors connected by a localspeed computer bus, which is a subsystem that transfers data between components Europe Parallel computing for computation cern Worldwide LHC Computing Grid (WLCG) 6

7 109 7

8 천리안카메라 무게 45kg, 가로/세로 75 cm, 높이 50 cm 로 일반 디지털카메라보다 100배 이상 선 명한 10억화소급 소형 카메라: 크기는 최소촬영범위는 최대 거리 사진 찍어 확대하면 사람/ 간판 등 또렷하게 보여 가격: 10만달러 도심촬영시 800m밖 우표구별, 광활한 지역 생태 연구시 장소 옮길 필요없음 데이터 처리 한계 때문에 흑백사진만 찍 을 수 있음 8

9 천리안카메라 vs. CMS 검출기 무게 45kg, 가로/세로 75 cm, 높이 50 cm 로 일반 디지털카메라보다 100배 이상 선 명한 10억화소급 소형 카메라: 크기는 최소촬영범위는 최대 거리 사진 찍어 확대하면 사람/ 간판 등 또렷하게 보여 가격: 10만달러 도심촬영시 800m밖 우표구별, 광활한 지역 생태 연구시 장소 옮길 필요없음 데이터 처리 한계 때문에 흑백사진만 찍 을 수 있음 무게 14,000톤, 길이 22 m, 직경 15 m로 지구 자기장의 100,000배 이상 큰 자기장을 이용한 초대형 입자 검출기: 크기는 최대검출기시스템은 콤팩트 (초대형 초전도 솔레노이드 전자석안에 설치됨) 강력한 자기장을 이용하면 입자의 궤적(운동량)을 정확하게 측정 가격: 모름 (LHC~100억달러) 우주생성의 비밀(새물리)을 밝히고 세 상을 바꿀 새로운 기술도 개발 CERN연구소내 데이터 처리 한계 9

10 Tier sites Tier-0: CERN Computer Centre Tier-1 (11 sites) Tier-2 (140 sites) universities and other scientific institutes Tier-3 (lots) local clusters in a university department or an individual PC Why tiered?

11 Tier functions Tier-0: CERN Computer Centre first accepts RAW data repacks the RAW data into primary datasets archives the RAW data to tape distributes into T1 (i.e, two copies) prompt calibration to get calibration constants prompt first pass reconstruction (RECO data) distributes into T1 T1 re-reconstruction, skimming, calibration distributes into other T1, CERN, T2 stores data T2 grid-based analysis Monte Carlo (MC) simulation T0: 데이터처리 1 T1: 데이터재처리및저장 2 T2: 데이터분석 3

12 Tier services by middleware Storage Element (SE) Data access protocols & interfaces Computing Element (CE) Job-manager and worker nodes Security User interface Information service Workload management Software components

13 데이터처리 데이터저장 데이터분석 데이터용량 It s time to talk about... 13

14 Volume of data produced at LHC 천 백만 십억 조 천조 14

15 천 백만 십억 조 Volume of data produced at LHC Size of interest 천조 천조분의일 f (femto) 15

16 PByte PBytes/sec PBytes/sec LHC/CMS Tiered Data System MBytes/sec 38 countries, 3000 scientists Tens of PetaBytes/yr ExaBytes in ~10 years 16

17 PBytes/sec PByte PBytes/sec trigger MBytes/sec LHC/CMS Tiered Data System 38 countries, 3000 scientists Tens of PetaBytes/yr ExaBytes in ~10 years 17

record of data transfer over 10,000 km between CERN and California, with a

18 GBytes per second A record of data on backup tape with a transfer rate of 1.1 GB/s for several hours ( ) = Recording a movie on DVD every 4 s A record of data transfer over 10,000 km between CERN and California, with a throughput of 2.38 GB/s for over an hour ( ) = Sending 200 DVD films in an hour

~2000 storage servers are used to store experiment/user data EOS project started in Apr 2010 to provide fast

19 Storage: CASTOR EOS CERN stores 12 million new files per month in the CASTOR. Expected to increase 3 times until 2015 reaching EB. ~2000 storage servers are used to store experiment/user data EOS project started in Apr 2010 to provide fast and reliable disk only storage technology events/yr 1 event = 1.6 MB 3.2 PB/yr from CMS 8.5 GB DVD DVD thickness = 1.2 mm How many DVDs? 3.2PB / 8.5GB 40k DVDs How tall? 40k DVDs 1.2mm 500m

20 Tier 2 in Korea CMS (KNU) ALICE (KISTI)

21 GRID Real Time Monitoring CERN KOREA 21

22 Bonus: Distributed analysis

23 Car Driving ( ) driving ( ) GPS

24 Data analysis ( ) analysing ( ) Configuration

25 Distributed analysis

26 Interesting! vs.

27 Thanks

0125_ 워크샵 발표자료_완성.key

0125_ 워크샵 발표자료_완성.key WordPress is a free and open-source content management system (CMS) based on PHP and MySQL. WordPress is installed on a web server, which either is part of an Internet hosting service or is a network host