Grid Lab About the need of 3 Tier storage 5/22/121CHEP 2012, The need of 3 Tier storage Dmitri Ozerov Patrick Fuhrmann CHEP 2012, NYC, May 22, 2012 Grid.

Slides:



Advertisements
Similar presentations
Storage Workshop Summary Wahid Bhimji University Of Edinburgh On behalf all of the participants…
Advertisements

ATLAS Tier-3 in Geneva Szymon Gadomski, Uni GE at CSCS, November 2009 S. Gadomski, ”ATLAS T3 in Geneva", CSCS meeting, Nov 091 the Geneva ATLAS Tier-3.
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.
Data Clustering Research in CMS Koen Holtman CERN/CMS Eindhoven University of technology CHEP ’2000 Feb 7-11, 2000.
15/07/2010Swiss WLCG Operations Meeting Summary of the last GridKA Cloud Meeting (07 July 2010) Marc Goulette (University of Geneva)
Outline Network related issues and thinking for FAX Cost among sites, who has problems Analytics of FAX meta data, what are the problems  The main object.
I/O Tanenbaum, ch. 5 p. 329 – 427 Silberschatz, ch. 13 p
Process Management A process is a program in execution. It is a unit of work within the system. Program is a passive entity, process is an active entity.
Hall D Online Data Acquisition CEBAF provides us with a tremendous scientific opportunity for understanding one of the fundamental forces of nature. 75.
Filesytems and file access Wahid Bhimji University of Edinburgh, Sam Skipsey, Chris Walker …. Apr-101Wahid Bhimji – Files access.
Lecture 11: DMBS Internals
Wahid Bhimji University of Edinburgh J. Cranshaw, P. van Gemmeren, D. Malon, R. D. Schaffer, and I. Vukotic On behalf of the ATLAS collaboration CHEP 2012.
LHCC Comprehensive Review – September WLCG Commissioning Schedule Still an ambitious programme ahead Still an ambitious programme ahead Timely testing.
ALICE data access WLCG data WG revival 4 October 2013.
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.
Alexandre A. P. Suaide VI DOSAR workshop, São Paulo, 2005 STAR grid activities and São Paulo experience.
Computing for ILC experiment Computing Research Center, KEK Hiroyuki Matsunaga.
José M. Hernández CIEMAT Grid Computing in the Experiment at LHC Jornada de usuarios de Infraestructuras Grid January 2012, CIEMAT, Madrid.
Data & Storage Services CERN IT Department CH-1211 Genève 23 Switzerland t DSS From data management to storage services to the next challenges.
LCG Service Challenge Phase 4: Piano di attività e impatto sulla infrastruttura di rete 1 Service Challenge Phase 4: Piano di attività e impatto sulla.
Data management for ATLAS, ALICE and VOCE in the Czech Republic L.Fiala, J. Chudoba, J. Kosina, J. Krasova, M. Lokajicek, J. Svec, J. Kmunicek, D. Kouril,
Your university or experiment logo here Caitriana Nicholson University of Glasgow Dynamic Data Replication in LCG 2008.
GStore: GSI Mass Storage ITEE-Palaver GSI Horst Göringer, Matthias Feyerabend, Sergei Sedykh
Data transfer over the wide area network with a large round trip time H. Matsunaga, T. Isobe, T. Mashimo, H. Sakamoto, I. Ueda International Center for.
14 Aug 08DOE Review John Huth ATLAS Computing at Harvard John Huth.
Evolution, by tackling new challenges| CHEP 2015, Japan | Patrick Fuhrmann | 16 April 2015 | 1 Patrick Fuhrmann On behave of the project team Evolution,
D C a c h e Michael Ernst Patrick Fuhrmann Tigran Mkrtchyan d C a c h e M. Ernst, P. Fuhrmann, T. Mkrtchyan Chep 2003 Chep2003 UCSD, California.
Introduction to dCache Zhenping (Jane) Liu ATLAS Computing Facility, Physics Department Brookhaven National Lab 09/12 – 09/13, 2005 USATLAS Tier-1 & Tier-2.
Status Report of WLCG Tier-1 candidate for KISTI-GSDC Sang-Un Ahn, for the GSDC Tier-1 Team GSDC Tier-1 Team 12 th CERN-Korea.
LCG Phase 2 Planning Meeting - Friday July 30th, 2004 Jean-Yves Nief CC-IN2P3, Lyon An example of a data access model in a Tier 1.
DYNES Storage Infrastructure Artur Barczyk California Institute of Technology LHCOPN Meeting Geneva, October 07, 2010.
1 Process migration n why migrate processes n main concepts n PM design objectives n design issues n freezing and restarting a process n address space.
ROOT and Federated Data Stores What Features We Would Like Fons Rademakers CERN CC-IN2P3, Nov, 2011, Lyon, France.
Analysis in STEP09 at TOKYO Hiroyuki Matsunaga University of Tokyo WLCG STEP'09 Post-Mortem Workshop.
BNL Wide Area Data Transfer for RHIC & ATLAS: Experience and Plans Bruce G. Gibbard CHEP 2006 Mumbai, India.
Caitriana Nicholson, CHEP 2006, Mumbai Caitriana Nicholson University of Glasgow Grid Data Management: Simulations of LCG 2008.
1 User Analysis Workgroup Discussion  Understand and document analysis models  Best in a way that allows to compare them easily.
CERN IT Department CH-1211 Genève 23 Switzerland t Frédéric Hemmer IT Department Head - CERN 23 rd August 2010 Status of LHC Computing from.
Summary of Data Management Jamboree Ian Bird WLCG Workshop Imperial College 7 th July 2010.
USATLAS dCache System and Service Challenge at BNL Zhenping (Jane) Liu RHIC/ATLAS Computing Facility, Physics Department Brookhaven National Lab 10/13/2005.
CMS Computing Model Simulation Stephen Gowdy/FNAL 30th April 2015CMS Computing Model Simulation1.
SLACFederated Storage Workshop Summary For pre-GDB (Data Access) Meeting 5/13/14 Andrew Hanushevsky SLAC National Accelerator Laboratory.
Scientific Storage at FNAL Gerard Bernabeu Altayo Dmitry Litvintsev Gene Oleynik 14/10/2015.
BNL Service Challenge 3 Status Report Xin Zhao, Zhenping Liu, Wensheng Deng, Razvan Popescu, Dantong Yu and Bruce Gibbard USATLAS Computing Facility Brookhaven.
Storage and Data Movement at FNAL D. Petravick CHEP 2003.
US-CMS T2 Centers US-CMS Tier 2 Report Patricia McBride Fermilab GDB Meeting August 31, 2007 Triumf - Vancouver.
Andrea Manzi CERN On behalf of the DPM team HEPiX Fall 2014 Workshop DPM performance tuning hints for HTTP/WebDAV and Xrootd 1 16/10/2014.
PROOF tests at BNL Sergey Panitkin, Robert Petkus, Ofer Rind BNL May 28, 2008 Ann Arbor, MI.
PROOF Benchmark on Different Hardware Configurations 1 11/29/2007 Neng Xu, University of Wisconsin-Madison Mengmeng Chen, Annabelle Leung, Bruce Mellado,
Data Placement Intro Dirk Duellmann WLCG TEG Workshop Amsterdam 24. Jan 2012.
Stephen Gowdy FNAL 9th Feb 2015CMS Computing Model Simulation 1.
1 A Scalable Distributed Data Management System for ATLAS David Cameron CERN CHEP 2006 Mumbai, India.
Latest Improvements in the PROOF system Bleeding Edge Physics with Bleeding Edge Computing Fons Rademakers, Gerri Ganis, Jan Iwaszkiewicz CERN.
BNL dCache Status and Plan CHEP07: September 2-7, 2007 Zhenping (Jane) Liu for the BNL RACF Storage Group.
Introduce Caching Technologies using Xrootd Wei Yang 10/28/14ATLAS TIM 2014 Univ. Chicago1.
LHCC Referees Meeting – 28 June LCG-2 Data Management Planning Ian Bird LHCC Referees Meeting 28 th June 2004.
Improving Performance using the LINUX IO Scheduler Shaun de Witt STFC ISGC2016.
IT-DSS Alberto Pace2 ? Detecting particles (experiments) Accelerating particle beams Large-scale computing (Analysis) Discovery We are here The mission.
Status of GSDC, KISTI Sang-Un Ahn, for the GSDC Tier-1 Team
1 Lecture 16: Data Storage Wednesday, November 6, 2006.
ATLAS TIER3 in Valencia Santiago González de la Hoz IFIC – Instituto de Física Corpuscular (Valencia)
Open Science Grid Consortium Storage on Open Science Grid Placing, Using and Retrieving Data on OSG Resources Abhishek Singh Rana OSG Users Meeting July.
Availability of ALICE Grid resources in Germany Kilian Schwarz GSI Darmstadt ALICE Offline Week.
Update on Plan for KISTI-GSDC
Ákos Frohner EGEE'08 September 2008
Lecture 11: DMBS Internals
Grid Canada Testbed using HEP applications
Chapter 2: Operating-System Structures
Chapter 2: Operating-System Structures
The LHCb Computing Data Challenge DC06
Presentation transcript:

Grid Lab About the need of 3 Tier storage 5/22/121CHEP 2012, The need of 3 Tier storage Dmitri Ozerov Patrick Fuhrmann CHEP 2012, NYC, May 22, 2012 Grid Lab Partially evaluated by

Grid Lab Profile of a “large” WLCG Tier 2. Efficiency and performance of Storage Media. Replaying billing information. Simple cache filling mechanism. Improved mechanism. Summary 5/22/12CHEP 2012, The need of 3 Tier storage2 Content

Grid Lab About the need of 3 Tier storage 5/22/123CHEP 2012, The need of 3 Tier storage Spinning Disk SSD Tertiary Storage (Tape) Spinning Disk Tertiary Storage (Tape)

Grid Lab 5/22/124CHEP 2012, The need of 3 Tier storage Our evaluation results are most relevant for storage providers of Tier 2/3. The WLCG 3 Tier Model Spinning Disk Tertiary Storage (Tape) Spinning Disk Tier 0 / 1 Tier 2

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage5 Activities of a large Tier 2 Most of the clients are using dCap. Data is read directly from storage. We found a significant difference in transfer speed between dCap and gridFTP. (From few MB/Sec to 30 MB/s due to different access pattern)

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage6 Partially reading files 870 TB of data was transferred over the network; files were only partially read. The transferred data corresponds to 24 PB, if the entire file would have been transferred over the wire. The average network bandwidth: 0.33 GB/sec -> 9 GB/sec Load on the storage moves from dedicated effective storage pools to local space on WN Is it more efficient to copy data to WN storage, to be processed locally ? CMS User group statistics : Nov 2011

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage7 Efficiency of storage hardware Investigations were made with the DESY Grid Lab facility (For details, please see our poster 503) Processing I/O intensive ATLAS Hammercloud jobs revealed two limitations: 50 Jobs / 100 TB – efficiency of jobs drops by 30% Running more than 80 Jobs / 100TB is a waste of CPU resources.

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage8 Storage media performance 1.Traditional Disk (HDD) – good in streaming, bad in random reads SSD -> 10 times faster in random reading 2.Copying the entire file to the WN instead of directly accessing the data will be affected significantly due to the increased number of reads.

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage9 SSD Spinning Disk Simple Strategy 1.Get file-A from disk 2.Asynchronously copy the same file to SSD storage, removing “least recently used” files in case space is running short. SSD Spinning Disk 1.Next read will get file-A directly from fast SSD storage as long as file is still present.

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage10 How was the analysis done ? DESY Storage Element Billing File Or Billing DB 9:00 transfer Records each single transfer with time, file size etc. Simulation Of Various scenarios

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage11 Result: 20 Tb Cache Size : 70% of traffic to clients is originated from SSD. 380TB was streamed in total (with 140 Mbyts/sec ) Re-play of Storage System billing info 1 st November 2011: The cache is completely empty Read file from cache if present If not -> read from disk: copy to cache if no space left, remove "least accessed" file. Concerning bytes transferred

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage12 Re-play of Storage System billing info Read file from cache if present If not -> read from disk: copy to cache if no space left, remove "least accessed" file. Result: 20 Tb Cache Size : 70% of traffic to clients is originated from SSD. 380TB was streamed in total (with 140 Mbyts/sec ) Only 4% of reads from SSD 1 st November 2011: The cache is completely empty Concerning # of reads

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage13 Agnostic against abnormal activity To prove that the evaluation has not been mislead by abnormal activities in the observed time frame (e.g. user code bugs or specific jobs only seen those months) an additional month was analyzed.

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage14 Results from a less active month With 20 TB of cache, 80% of reads are redirected to the cache

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage15 SSD Spinning Disk Improved Strategy 1.Get file-A from disk 2.Read the file up to 10 times 3.Asynchronously copy the same file to SSD storage with the 10 th read SSD Spinning Disk 1.Next read will get file-A directly from fast SSD storage Copy with 10 th access

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage16 Copy to SSD cache after 10 file accesses With the improved (tunable) algorithm, we are able to even reduce data transfer, while keeping high cache access rates.

Grid Lab 5/22/12CHEP 2012, The need of 3 Tier storage17 Summary Storage systems based on traditional spinning disks have a limited I/O performance. Adding SSDs, as an additional cache layer in currently deployed storage solutions (DPM, dCache..) can boost I/O performance significantly. An analysis of CMS user activities has shown that only 10% of cache, results in 70% of random reads from the cache. A storage system with 3 Tiers is transparent for the user and cost effective for sites.