Performance tests of storage arrays Irina Makhlyueva ALICE DAQ group 20 September 2004.

Slides:



Advertisements
Similar presentations
Enhanced Availability With RAID CC5493/7493. RAID Redundant Array of Independent Disks RAID is implemented to improve: –IO throughput (speed) and –Availability.
Advertisements

Linux Filesystem Features Evolution of a de facto standard file system for Linux: ‘ext2’
Denny Cherry Manager of Information Systems MVP, MCSA, MCDBA, MCTS, MCITP.
NWfs A ubiquitous, scalable content management system with grid enabled cross site data replication and active storage. R. Scott Studham.
Linux Filesystem Features Evolution of a de facto standard file system for Linux: ‘ext2’
1 CS143: Disks and Files. 2 System Architecture CPU Main Memory Disk Controller... Disk Word (1B – 64B) ~ x GB/sec Block (512B – 50KB) ~ x MB/sec System.
Silberschatz, Galvin and Gagne ©2009 Operating System Concepts – 8 th Edition Mass-Storage Systems Revised Tao Yang.
Introduction to Database Systems 1 The Storage Hierarchy and Magnetic Disks Storage Technology: Topic 1.
Secondary Storage Unit 013: Systems Architecture Workbook: Secondary Storage 1G.
Disk Array Performance Estimation AGH University of Science and Technology Department of Computer Science Jacek Marmuszewski Darin Nikołow, Marek Pogoda,
CERN IT Department CH-1211 Genève 23 Switzerland t R&D Activities on Storage in CERN-IT’s FIO group Helge Meinhard / CERN-IT HEPiX Fall 2009.
Sponsored by: PASS Summit 2010 Preview Storage for the DBA Denny Cherry MVP, MCSA, MCDBA, MCTS, MCITP.
Database Services for Physics at CERN with Oracle 10g RAC HEPiX - April 4th 2006, Rome Luca Canali, CERN.
Tier 1A Storage Procurement 2001/2002 Andrew Sansum CLRC eScience Centre.
Terabyte IDE RAID-5 Disk Arrays David A. Sanders, Lucien M. Cremaldi, Vance Eschenburg, Romulus Godang, Christopher N. Lawrence, Chris Riley, and Donald.
1 A Basic R&D for an Analysis Framework Distributed on Wide Area Network Hiroshi Sakamoto International Center for Elementary Particle Physics (ICEPP),
File System Benchmarking
Parity Logging O vercoming the Small Write Problem in Redundant Disk Arrays Daniel Stodolsky Garth Gibson Mark Holland.
Guide to Linux Installation and Administration, 2e 1 Chapter 9 Preparing for Emergencies.
Optimizing Performance of HPC Storage Systems
DAC-FF The Ultimate Fibre-to-Fibre Channel External RAID Controller Solution for High Performance Servers, Clusters, and Storage Area Networks (SAN)
Storage Systems Market Analysis Dec 04. Storage Market & Technologies.
CASPUR Site Report Andrei Maslennikov Lead - Systems Karlsruhe, May 2005.
CASPUR / CERN / CSP / DataDirect / Panasas / RZ Garching Im Storage etwas Neues (Not All is Quiet on the Storage Front) Andrei Maslennikov CASPUR Consortium.
Sensitivity of Cluster File System Access to I/O Server Selection A. Apon, P. Wolinski, and G. Amerson University of Arkansas.
Tape Storage Performance Jonathan Maye-Hobbs William Buehler 2011 Computer System, Cluster, and Networking Summer Institute.
Jefferson Lab Site Report Kelvin Edwards Thomas Jefferson National Accelerator Facility Newport News, Virginia USA
Amy Apon, Pawel Wolinski, Dennis Reed Greg Amerson, Prathima Gorjala University of Arkansas Commercial Applications of High Performance Computing Massive.
Linux Servers with JASMine K. Edwards, A. Kowalski, S. Philpott HEPiX May 21, 2003.
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.
Wide Area Network Access to CMS Data Using the Lustre Filesystem J. L. Rodriguez †, P. Avery*, T. Brody †, D. Bourilkov *, Y.Fu *, B. Kim *, C. Prescott.
Experience with the Thumper Wei Yang Stanford Linear Accelerator Center May 27-28, 2008 US ATLAS Tier 2/3 workshop University of Michigan, Ann Arbor.
SLAC Site Report Chuck Boeheim Assistant Director, SLAC Computing Services.
CASPUR Storage Lab Andrei Maslennikov CASPUR Consortium Catania, April 2002.
10/22/2002Bernd Panzer-Steindel, CERN/IT1 Data Challenges and Fabric Architecture.
Optimisation of Grid Enabled Storage at Small Sites Jamie K. Ferguson University of Glasgow – Jamie K. Ferguson – University.
CASPUR Site Report Andrei Maslennikov Lead - Systems Amsterdam, May 2003.
CASPUR / GARR / CERN / CNAF / CSP New results from CASPUR Storage Lab Andrei Maslennikov CASPUR Consortium May 2003.
Forschungszentrum Karlsruhe in der Helmholtz-Gemeinschaft Implementation of a reliable and expandable on-line storage for compute clusters Jos van Wezel.
RAL Site Report John Gordon HEPiX/HEPNT Catania 17th April 2002.
Status SC3 SARA/Nikhef 20 juli Status & results SC3 throughput phase SARA/Nikhef Mark van de Sanden.
Sep. 17, 2002BESIII Review Meeting BESIII DAQ System BESIII Review Meeting IHEP · Beijing · China Sep , 2002.
Tier-2 storage A hardware view. HEP Storage dCache –needs feed and care although setup is now easier. DPM –easier to deploy xrootd (as system) is also.
Linux IDE Disk Servers Andrew Sansum 8 March 2000.
PROOF tests at BNL Sergey Panitkin, Robert Petkus, Ofer Rind BNL May 28, 2008 Ann Arbor, MI.
COMPASS Computerized Analysis and Storage Server Iain Last.
SA1 operational policy training, Athens 20-21/01/05 Presentation of the HG Node “Isabella” and operational experience Antonis Zissimos Member of ICCS administration.
Tested, seen, heard… Andrei Maslennikov Rome, April 2006.
01. December 2004Bernd Panzer-Steindel, CERN/IT1 Tape Storage Issues Bernd Panzer-Steindel LCG Fabric Area Manager CERN/IT.
CERN IT Department CH-1211 Genève 23 Switzerland t The Tape Service at CERN Vladimír Bahyl IT-FIO-TSI June 2009.
B ENCHMARK ON D ELL 2950+MD1000 ATLAS Tier2/Tier3 workshop Wenjing wu AGLT2 / University of Michigan 2008/05/27.
AMS02 Data Volume, Staging and Archiving Issues AMS Computing Meeting CERN April 8, 2002 Alexei Klimentov.
Improving Performance using the LINUX IO Scheduler Shaun de Witt STFC ISGC2016.
Enhanced Availability With RAID CC5493/7493. RAID Redundant Array of Independent Disks RAID is implemented to improve: –IO throughput (speed) and –Availability.
CRISP WP18, High-speed data recording Krzysztof Wrona, European XFEL PSI, 18 March 2013.
PC COMPONENTS. System Unit Cases This is the cabinet that holds the main components of a computer. It includes a plastic front panel for aesthetic purpose.
RAID TECHNOLOGY RASHMI ACHARYA CSE(A) RG NO
CERN Disk Storage Technology Choices LCG-France Meeting April 8 th 2005 CERN.ch.
29/04/2008ALICE-FAIR Computing Meeting1 Resulting Figures of Performance Tests on I/O Intensive ALICE Analysis Jobs.
Bernd Panzer-Steindel CERN/IT/ADC1 Medium Term Issues for the Data Challenges.
15.June 2004Bernd Panzer-Steindel, CERN/IT1 CERN Mass Storage Issues.
Ryan Leonard Storage and Solutions Architect
Fujitsu Training Documentation RAID Groups and Volumes
Bernd Panzer-Steindel, CERN/IT
Experience of Lustre at a Tier-2 site
Pierre-Emmanuel Brinette
VNX Storage Report Project: Sample VNX Report Project ID:
Oracle Storage Performance Studies
Unity Storage Array Profile
ASM-based storage to scale out the Database Services for Physics
Presentation transcript:

Performance tests of storage arrays Irina Makhlyueva ALICE DAQ group 20 September 2004

I.Makhlyueva - ALICE DAQ Meeting2 20 September 2004 Contents Goals Tested storage array types RAID performance tests  Early tests  Tests at CASPUR Storage Lab  Performance tests with Infortrend and dotHill RAIDs Summary, final remarks

I.Makhlyueva - ALICE DAQ Meeting3 20 September 2004 Goals RAID systems in ALICE DAQ:  temporary data buffer in case of CDR failures  main file storage in the Reference DAQ system We have tested: Different brands and models of RAIDs Performance of different file systems Influence of RAID parameters Performance for simultaneous multiple i/o operations

I.Makhlyueva - ALICE DAQ Meeting4 20 September 2004 Tested storage arrays Infortrend IFT-6330  12 IDE drive slots  128 MB cache  Dual Gbps fiber host channel Infortrend EonStor A16F-G1A2 (CASPUR SLAB)  16 SATA drive slots  Up to 1GB cache  Dual 2Gbps fiber host channel dotHILL SANnet II 200 FC  12 fiber channel disk slots (expandable)  1 GB cache  2 Gbps fiber host channel

I.Makhlyueva - ALICE DAQ Meeting5 20 September 2004 Performance tests The simplest test: filling an empty RAID with fixed-size files, measure the average transfer rate (~4 h per test). % of full disk MB/s MB/s % of full disk Large fluctuationsLack of repeatablity from test to test Sudden “jumps” within one test Infortrend IFT-6330 (1.1 TB) with ext3 fs Standard CERN Linux PC (RH 7.3) Unexpected behaviour was observed

I.Makhlyueva - ALICE DAQ Meeting6 20 September 2004 Tests at CASPUR Storage Lab The problem was investigated in collaboration with CERN IT and CASPUR (Rome), using storage arrays (of different brands) of the CASPUR Slab. The main focus was on (see Ref. [1]): filesystem type and its parameters fs mount options kernel tuning RAID parameter tuning (stripe size) Main results: transfer rate “jumps”, reproducibility problems: cured by kernel and ext3 tuning xfs gives a substantially better performance than ext3 bugs in Infortrend RAID firmware: acknowledged by the firm

I.Makhlyueva - ALICE DAQ Meeting7 20 September K 128k 256K EXT3 results – filling 1.7 TB with 8 GB files Much smaller fluctuations A firmware problem became visible: a dependence on RAID-5 stripe size Infortrend SATA results

I.Makhlyueva - ALICE DAQ Meeting8 20 September 2004 filesystem type dependence XFS higher performance better stability less sensitive to firmware flaws Infortrend SATA results

I.Makhlyueva - ALICE DAQ Meeting9 20 September 2004 Further tests at ALICE DAQ lab Most up-to-date version of XFS (special Linux RedHat 9 installation) Two RAID systems mounted simultaneously, via a Brokade FC switch:  Infortrend IFT-6330  High-end dotHILL SANnet II 200 FC Dependence of the file size and record length Concurrent i/o: single or multiple “writer” or/and “reader” process(es) “Pile-up” (random reading) tests % of full disk MB/s IFT, write IFT, read DH, write DH, read 2GB, 8MB

I.Makhlyueva - ALICE DAQ Meeting10 20 September 2004 Dependence on the file size and record length 2 GB, write dothill 1 dothill 2 dothill 1 2 GB, read IFT 1 IFT 2 2 GB, write IFT 2 IFT 1 2 GB, read for both RAID systems, the i/o rates were measured for filesizes of 100, 300, 1024 and 2048 MB and recl=8 kB, 32 kB, 128 kB, 512 kB, 2 MB, 8 MB, 32 MB and 128 MB Example (fsize=2 GB):

I.Makhlyueva - ALICE DAQ Meeting11 20 September 2004 Concurrent i/o “diskperf” benchmark (fsize=2 GB, recl=8MB) Aggregate writing transfer rate: extra writer(s): no (dotHill) or weak (IFT) effect extra reader(s): very strong effect in case of IFT (firmware?) a peculiar feature: reading speed is strongly suppressed in presence of writing process(es) – both dotHill and IFT writer(s) only writers + reader(s)

I.Makhlyueva - ALICE DAQ Meeting12 20 September 2004 Concurrent i/o “diskperf” benchmark (fsize=2 GB, recl=8MB) Aggregate writing transfer rate: extra writer(s): no (dotHill) or weak (IFT) effect extra reader(s): very strong effect in case of IFT (firmware?) a peculiar feature: reading speed is strongly suppressed in presence of writing process(es) – both dotHill and IFT writer(s) only writers + reader(s)

I.Makhlyueva - ALICE DAQ Meeting13 20 September 2004 Observation: with concurrent processes, the writer(s) dominate over the readers

I.Makhlyueva - ALICE DAQ Meeting14 20 September 2004 Random reading (“pile-up”) test One of standard storage performance benchmarks used at CERN is a “pile- up test” by R.Töbicke: a multiple-thread random access reading over an arbitrary number of 2 GB files, prepared in advance. Our results: the aggregate reading speed is only a small fraction of the maximal reading speed for serial access it depends on the number of files used

I.Makhlyueva - ALICE DAQ Meeting15 20 September 2004 Summary a stable RAID performance can be achieved by a careful tuning of the Linux kernel and the file system parameters XFS is better than EXT3 IFT RAID: aggregate writing speed is degraded by a presence of concurrent reading processes (a firmware effect?) for both tested RAID systems the reading speed is sharply suppressed in presence of concurrent writing process(es) dotHill system features a superior performance but is extremely expensive. Further tests may include: repeat the tests on the IFT system after upgrading the firmware try different benchmarks (lmdd, iozone, …) study the performance in a multi-host environment

I.Makhlyueva - ALICE DAQ Meeting16 20 September 2004 References Tests at CASPUR Slab: A.Maslennikov, New results from CASPUR Storage Lab, Thanks for help and discussions are to: CASPUR: A.Maslennikov CERN IT: A.Horvath, J.Iven, P.Kelemen, R.Többicke ALICE DAQ group: K.Schossmaier, P. Van De Vyver