1 a basic r&d for an analysis framework distributed on wide area network hiroshi sakamoto...
TRANSCRIPT
11
A Basic R&D for an Analysis A Basic R&D for an Analysis Framework Distributed on Wide Framework Distributed on Wide
Area NetworkArea Network
Hiroshi SakamotoHiroshi SakamotoInternational Center for Elementary Particle Physics International Center for Elementary Particle Physics
(ICEPP), the University of Tokyo(ICEPP), the University of TokyoForFor
Regional Center R&D Collaboration,Regional Center R&D Collaboration,Computing Research Center, KEKComputing Research Center, KEK
ICEPP, the University of TokyoICEPP, the University of Tokyo
22
KEK-ICEPP R&D CollaborationKEK-ICEPP R&D Collaboration
ICEPP, Univ. TokyoICEPP, Univ. TokyoRegional Analysis Center for LHC ATLAS ExperimentRegional Analysis Center for LHC ATLAS Experiment
KEKKEKCentral HEP Facility in JapanCentral HEP Facility in Japan
Joint R&D ProgramJoint R&D ProgramComputing Research Center, KEK and ICEPP, TokyoComputing Research Center, KEK and ICEPP, Tokyo
Started in 2001Started in 2001
For Distributed Analysis FrameworkFor Distributed Analysis Framework
Introduces Computing Grid TechnologyIntroduces Computing Grid Technology
33
R&D ItemsR&D ItemsPC FarmPC Farm
Blade ServersBlade Servers
Storage SystemStorage SystemIDE File Server PrototypeIDE File Server Prototype
Wide Area NetworkWide Area NetworkKEK-ICEPP GbE ConnectionKEK-ICEPP GbE Connection
CERN-ICEPP Connectivity StudyCERN-ICEPP Connectivity Study
Integrated StudyIntegrated StudyHPSS Access via WANHPSS Access via WAN
44
HP Blade ServerHP Blade ServerHP ProLiant BL20p HP ProLiant BL20p
Xeon 2.8GHz 2CPU/nodeXeon 2.8GHz 2CPU/node
1GB memory1GB memory
SCSI 36GBx2, hardware RAID1SCSI 36GBx2, hardware RAID1
3 GbE NIC3 GbE NIC
iLO remote administration tooliLO remote administration tool
8 blades/ Enclosure(6U)8 blades/ Enclosure(6U)
Total 108 blades(216 CPUs) in 3 racksTotal 108 blades(216 CPUs) in 3 racks
Rocks 2.3.2(RedHat 7.3 base) for Cluster managRocks 2.3.2(RedHat 7.3 base) for Cluster managementement
55
HP Blade Server 2HP Blade Server 2
66
HP Blade Server Network HP Blade Server Network ConfigurationConfiguration
SW
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
SW
SW
FrontendServer
CampusLAN
Private LANPXE/Control
Private LANData
Private LANiLO
Blade Servers
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
NIC1
NIC2
NIC3
iLO
77
File Server PrototypeFile Server Prototype
2 Xeon 2.8GHz (FSB 533MHz)2 Xeon 2.8GHz (FSB 533MHz)
Memory 1GBMemory 1GB
2 PCI RAID I/F, 3ware 7500-82 PCI RAID I/F, 3ware 7500-8
2 GbE(1000Base-T) LAN2 GbE(1000Base-T) LAN
4U Rack mount Case, 16 hot-swap IDE HDD bay4U Rack mount Case, 16 hot-swap IDE HDD bay
16 250GB HDD + 40GB HDD16 250GB HDD + 40GB HDD
460W N+1 Redundant Power Supply460W N+1 Redundant Power Supply
7 servers, total 24.5TB7 servers, total 24.5TB
88
8 HDD RAID5
0
20
40
60
80
100
120
140
160
180
EXT2 EXT3 XFS J FS ReiserFS
FileSystem
Tra
nsf
er
Rat
e (
MB
/s)
READ WRITE
File Server Prototype 2File Server Prototype 2BenchmarksBenchmarks
Read 350MB/s, Read 350MB/s, Write 170-250MB/s Write 170-250MB/s RAID0 16 HDDRAID0 16 HDD
Read Read 140-160MB/s, 140-160MB/s, Write 20-50MB/s, Write 20-50MB/s, RAID5 8 HDDRAID5 8 HDD
16 HDD RAID0
0
50
100
150
200
250
300
350
400
EXT2 EXT3 XFS J FS ReiserFS
FileSystem
Tra
nsf
er
Rat
e (
MB
/s)
READ WRITE
RAID0 16HDD
RAID5 8 HDD
99
GbE between KEK and ICEPPGbE between KEK and ICEPP10Gbps SuperSINET Backbone10Gbps SuperSINET Backbone
CommodityCommodity
Project Oriented 1GbpsProject Oriented 1GbpsKEK-ICEPP for ATLAS ExperimentKEK-ICEPP for ATLAS Experiment
1010
WAN Performance MeasurementWAN Performance Measurement“A” settingTCP 479Mbps -P 1 -t 1200 -w 128KBTCP 925Mbps -P 2 -t 1200 -w 128KBTCP 931Mbps -P 4 -t 1200 -w 128KBUDP 953Mbps -b 1000MB -t 1200 -w 128KB
"A" setting: 104.9MB/s "B" setting: 110.2MB/s
“B” settingTCP 922Mbps -P 1 -t 1200 -w 4096KBUDP 954Mbps -b 1000MB -t 1200 -w 4096KB
1111
Wide Area NetworkWide Area NetworkInternational networkInternational network
2.5Gbps x 2 (=5Gbps) to NY2.5Gbps x 2 (=5Gbps) to NY
1212
CERN-Tokyo (Very preliminary)CERN-Tokyo (Very preliminary)[UDP] 17:00-17:30option: -b 1000MB -w 256KB[ ID] Interval Transfer Bandwidth Jitter Lost/Total Datagrams [ 3] 0.0-1095.5 sec 116 GBytes 906 Mbits/sec 0.015 ms 4723300/89157875 (5.3%)[ 3] 0.0-1095.5 sec 134933 datagrams received out-of-order
[TCP] 11:00-16:30[ ID] Interval Transfer Bandwidth Option[SUM] 0.0-1202.7 sec 645 MBytes 4.50 Mbits/sec -P 4 -w 256KB[SUM] 0.0-1203.3 sec 1.13 GBytes 8.04 Mbits/sec -P 8 -w 256KB[SUM] 0.0-1215.0 sec 2.35 GBytes 16.6 Mbits/sec -P 16 -w 256KB[SUM] 0.0-1219.9 sec 5.17 GBytes 36.4 Mbits/sec -P 32 -w 256KB[SUM] 0.0-1230.9 sec 8.30 GBytes 57.9 Mbits/sec -P 64 -w 256KB[SUM] 0.0-1860.0 sec 30.2 GBytes 140 Mbits/sec -P 128 -w 256KB[SUM] 0.0-1860.2 sec 37.0 GBytes 171 Mbits/sec -P 160 -w 256KB
0
20
40
60
80
100
120
140
4 8 16 32 64 128
1313
1Gbps
100Mbps
ICEPP KEK
100 CPUs6CPUs
HPSS 120TB
GRID testbed environmentwith HPSS through GbE-WAN
NorduGrid - grid-manager - gridftp-serverGlobus-mdsGlobus-replicaPBS server
NorduGrid - grid-manager - gridftp-serverGlobus-mdsPBS server
PBS clientsPBS clients
HPSS servers
~ 60km0.2TBSECE
CECE
SE
User PCs
1414
Basic Network Performance Basic Network Performance
0 0.2 0.4 0.6 0.8 1[106]
0
200
400
600744Mb/s 2.8ms
700Mb/s 0.39ms
HPSS mover -> KEK client HPSS mover -> ICEPP client
HPSS mover Bufsize=256kB
Network performance by netperf
TCP Buffer size of Client (Byte)
Transfe
r speed (M
Bit/s)
RTT~3msRTT~3ms
packet loss free.packet loss free.
MTU=1500MTU=1500
CPU/NIC is CPU/NIC is bottleneck.bottleneck.
Max TCP Buffer Max TCP Buffer Size(256k) in Size(256k) in HPSS servers HPSS servers cannot be cannot be changed.changed.(optimized for (optimized for IBM SP switch)IBM SP switch)
WANLAN
1515
Client disk speed @ KEK = 48MB/sClient disk speed @ ICEPP=33MB/s
0 2 4 6 8 100
20
40
60
80
KEK client (LAN) ICEPP client(WAN)
# of file transfer in parallel
Agg
rega
te T
rans
fer
spee
d (M
B/s
)Pftp→pftp HPSS mover disk Client disk
client disk speed 35-45MB/s
to /dev/null
to client disk Ftp buffer=64MB
1616
SummarySummary
R&D for Distributed Analysis FrameworkR&D for Distributed Analysis FrameworkStudy of FabricStudy of Fabric
Blade ServersBlade Servers
Disk Array/File Server PrototypeDisk Array/File Server Prototype
Network ConnectivityNetwork ConnectivityKEK-ICEPPKEK-ICEPP
CERN-ICEPPCERN-ICEPP
Integrated PerformanceIntegrated PerformanceHPSS Access via WANHPSS Access via WAN
1717
ProspectsProspects
Network ConnectivityNetwork ConnectivityCERN-KEK/Tokyo to 10Gbps in 2004CERN-KEK/Tokyo to 10Gbps in 2004
Tokyo-Taipei Connectivity Study soonTokyo-Taipei Connectivity Study soon
File Server UpgradeFile Server Upgrade100TB Disk Capacity100TB Disk Capacity
Hierarchical Storage Study soonHierarchical Storage Study soon
Prepare for ExperimentsPrepare for Experiments2004 Data Challenges2004 Data Challenges