Upgrade of KEK Central Computing System and Grid service - Indico
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Upgrade of KEK Central Computing System and Grid service T. Nakamura, G. Iwai, S. Kaneko, K. Murakami, T. Sasaki, S. Suzuki, W. Takase Computing Research Center Applied Research Laboratory HIGH ENERGY ACCELERATOR RESEARCH ORGANIZATION, KEK 2021-03-17 Tomoaki Nakamura, KEK-CRC 1
SuperKEKB: e+e- intensity frontier The first collision (Apr. 26th, 2018) Y. Yusa Design Luminosity: 8x1035 cm-2s-1 factor 40 higher than the previous KEKB H. Ikeda 2021-03-17 Tomoaki Nakamura, KEK-CRC 2
J-PARC: proton intensity frontier Fast extraction Slow extraction 2021-03-17 Tomoaki Nakamura, KEK-CRC 3
KEKCC: KEK Central Computing System Launched in Sep. 2016. No major upgrade in terms of the HWs during the 4 years operation period. Quite a stable phase. 54.55 Gbit/s K. Murakami 2021-03-17 Tomoaki Nakamura, KEK-CRC 4
Breakdown of CPU consumption Compute node Storage CPU: Intel Xeon E5-2697v3 (2.6GHz, 14cores) x 2 Disk: 10PB (GPFS, IBM ESS x8 racks) 358 nodes, 10,024 cores, 236kHS06/site 3PB (HSM cache) Memory: 4GB/core (8,000 cores) Interconnect: InfiniBand 4xFDR (54.55 Gbps) 8GB/core (2,000 cores) Tape: 70 PB (max cap.) CPU usage: breakdown by groups, normalized by the total CPU usage per month . Throughput 100 GB/s (Disk, GPFS), 50 GB/s (HSM, GHI) CPU usage has been reached more than 90 % of total resource. Incl. Belle/Belle II Belle II mainly G. Iwai 2021-03-17 Tomoaki Nakamura, KEK-CRC 5
Grid Jobs and Data Grid Jobs 168M HS06 hour/month (23.5 HS06/core) G. Iwai Grid storage read/write external data transfer 17 PB delivered to the other sites, 4 PB transferred to KEKCC during the 4 years. 2021-03-17 Tomoaki Nakamura, KEK-CRC 6
Schedule of procurement and construction Jan. 1st, 2019: Starting KEKCC specification formulation committee based on the Agreement on Government Procurement Schedule was determined Members from IPNS, IMSS, ACCL, ARL --- Installation manual --- Jun. 24th, 2019: Deadline for providing materials --- Draft specification --- Oct. 18th, 2019: Deadline for submission of comment to the draft specification --- Final specification --- Dec. 18th, 2019: Deadline for tender submission --- Technical evaluation --- Dec. 24th, 2019: Bid opening --- Execute a contract --- Jan. - Mar. 2020: Hardware delivery Apr. - Jul. 2020: System construction and setup Aug. 2020: Data migration and System stress test Sep 1st. 2020: Start operation 2021-03-17 Tomoaki Nakamura, KEK-CRC 7
Specification of KEKCC 2020 Supporting a lot of KEK projects, e.g., Belle/Belle2, ILC, various experiments in J-PARC, and so on. Rental system: KEKCC is entirely replaced every 4-5 years. Current KEKCC has started in September 2020 and will be ended in August 2024 or perhaps later. Data Analysis System Login servers, batch servers Lenovo ThinkSystem SD530, Intel Xeon Gold 6230 2.1 GHz, 283 kHS06 with 15,200 cores (40 cores x 380 nodes) Linux Cluster (CentOS 7.7) + LSF (job scheduler) Storage System IBM Elastic Storage System: 17 PB for GPFS + 8.5 PB for HSM cache (25.5 PB) HPSS: IBM TS4500 tape library (100 PB max.) Tape drive: TS1160 x72 Storage interconnect : IB 4xEDR Grid SE (StoRM) and iRODS access to GHI Total throughput : 100+ GB/s (Disk, GPFS) 60+ GB/s (HSM, GHI) Grid Computing System: UMD/EGI and iRODS/RENCI General-purpose IT Systems: mail, web (Indico, wiki, document archive), CA as well. 2021-03-17 Tomoaki Nakamura, KEK-CRC 9
Resource comparison Launched on Sep. 1st K. Murakami Upgrade 2016 2020 Factor CPU Xeon E5-2697v3 Haswell Xeon Gold 6230 Cascade Lake (2.6GHz, 14cores) (2.1 GHz, 20 cores) CPU cores 10,024 15,200 x1.5 HS06 236k 283k x1.2 OS SL 6.10 CentOS 7.7 Disk Capacity 10 + 3 PB (HSM) 17 + 8.5 PB (HSM) x2 Tape Drive IBM TS1150 x54 IBM TS1160 x72 Tape Media 7 TB/vol (JC) 7 TB /vol (JC) 10 TB/vol (JD), 360 MB/s 15 TB/vol (JD-Gen6) 20 TB/vol (JE), 400 MB/s Tape max 70 PB 100 PB x1.4 capacity Worker node configuration • CPU: 40 cores / node (18.63 HS06/core) • Memory: 4.8GB/core (304 nodes), 9.6GB/core (72 nodes) • Storage: 960GB SATA SSD / node 2021-03-17 Tomoaki Nakamura, KEK-CRC 10
Site scale evolution 2021-03-17 Tomoaki Nakamura, KEK-CRC 11
HW specification of Grid instances Type A: Head node of Grid instance x40 CPU: Xeon Gold 6230 2.10GHz 20Cores x1 Memory: 128GB Type A Type B Disk: 960GB SATA-SSD x2 x40 x8 Network: 1GE x2, 10GE x2, 16GFC x2, IB (4xEDR, 100Gbps) x1 Type B: Data transfer node x8 CPU: Xeon Gold 6230 2.10GHz 20Cores x1 Memory: 128GB Disk: 960GB SATA-SSD x2 Network: 1GE x2, 10GE x2, 40GE x1, IB (4xEDR, 100Gbps) x1 Type A BDII-top (AA), BDII-site (AA, UP) VOMS (HA, UP) LFC-Belle-RW (HA, UP), LFC-Belle-RO (AA, UP), LFC-Other (UP) SE-StoRM-FE-Belle-Raw (AA) SE-StoRM-BE-Belle-Raw Storage for Grid Service SE-StoRM-Belle-Ana (AA) Type B SE-StoRM-Other DTN-Belle-Raw (40GE x4) CE-ARC (AA) DTN-Belle-Ana (40GE x2) APEL DTN-Other (40G x2) • Almost service instances are deployed by CentOS7. CVMFS-Stratum0 (AS) • Some light-weight instances are running on KVM. CVMFS-Update • StoRM, VOMS, LFC are deployed on RHEL6. CVMFS-Stratum1 (AS) FTS (AA, UP) • HA: High Availability • CentOS7 packaging is not in time for the construction. • AA: Active-Active • OS Support will be terminated by the end of November. AMGA (AA, UP) • AS: Active-Standby • Use End of Lifecycle Support add-on (not-free). HTTP-Proxy (AA) • UP: External power supply ARGUS (AA, UP) Nagios 2021-03-17 Tomoaki Nakamura, KEK-CRC 12
Experiments dedicated servers Management Type A Type B Type C Belle II GPFS Server Front-End Server x1 x22 x14 x18 x2 x10 4xEDR x2 Belle II ESS Management x1 GHI mount (5.5PB) B II B II Grid Clients DIRAC etc. Belle II Temporal Storage Independent GPFS Storage Data capacity: 80TB RAID6 x3 = 240TB 3 days buffer: 700TB Backup capacity: 72TB RAID6 x2 = 144TB Storage for Belle II V5030 Raw Data Transfer Type A: Belle II x16 Type C: for Belle II x18 CPU: Xeon Gold 6230 2.10GHz 20Cores x1 CPU: Xeon Gold 6230 2.10GHz 20Cores x1 Memory: 128GB Memory: 128GB Disk: 2.4TB SAS-HDD x2 Disk: 1.6TB SAS-SSD x2 Network: 1GE x2, 10GE x2 Network: 1GE x2, 10GE x2, IB (4xEDR, 100Gbps) x1 Type B: Belle II x14 Power: External power supply (14 servers) CPU: Xeon Gold 6230 2.10GHz 20Cores x1 Front-End Server: x 10 Memory: 128GB CPU: Xeon Gold 6230 2.10GHz 20Cores x1 Disk: 2.4TB SAS-HDD x2 Memory: 128GB Network: 1GE x2, 10GE x2, 16GFC x2 Disk: 300GB SAS-HDD x2 Power: External power supply (10 servers) Network: 1GE x2, 10GE x2, IB (4xEDR, 100Gbps) x1, 10GE-F-DAQ-NW x2 2021-03-17 Tomoaki Nakamura, KEK-CRC 13
Summary of Grid service instance 2021-03-17 Tomoaki Nakamura, KEK-CRC 14
Concept of Grid system • Basic configuration does not change from the current system in terms of redundancy and robustness. Redundant configuration CE, CVMFS Stratum0/1, HTTP proxy, BDII-top, GridFTP servers behind StoRM High availability configuration by LifeKeeper VOMS, AMGA, LFC Uninterruptible operation against the scheduled power outage VOMS, AMGA, LFC, FTS3, ARGUS, BDII-site • All of the systems are built based on Centos 7 (partially RHEL8). • Some lightweight services are prepared by the virtual machine (KVM). • CREAM computing element are replaced to ARC-CE. • The capability of the data transfer nodes are strengthened. • Belle II Raw: 10Gx4 cables x 2 nodes (80G) → 40G x 4 nodes (160G) • Belle II Analysis: 10Gx2 cables x 2 nodes (40G) → 40G x 2 nodes (80G) • Other VO: 10Gx2 cables x 2 nodes (40G) → 40G x 2 nodes (80G) Belle II Raw Belle II Analysis Other VOs KEKCC 2016 G. Iwai 2021-03-17 Tomoaki Nakamura, KEK-CRC 15
Data transfer nodes for Belle II 100 Gbit/s 2021-03-17 Tomoaki Nakamura, KEK-CRC 16
Status of the new KEKCC Snapshot taken on Oct. 6th Max: 15200 Running jobs Pending jobs GPFS Read GPFS Write HSM Read HSM Write Designed value of the total transfer throughput • Disk only GPFS: 100GB/s • HSM and GHI: 50GB/s 2021-03-17 Tomoaki Nakamura, KEK-CRC 17
CPU consumption at KEKCC 2020 2021-03-17 Tomoaki Nakamura, KEK-CRC 18
Stored data in HSM Snapshot taken on Nov. 3rd K. Murakami 2021-03-17 Tomoaki Nakamura, KEK-CRC 19
Current Belle II raw data • The raw data amount for 50 ab-1 doesn’t correspond to 1 EB. • We expect the high-level trigger will be turned on in near future. • Fair-practical estimation: less than 100 PB for 50 ab-1 in 2029 We are here as of today 2021-03-17 Tomoaki Nakamura, KEK-CRC 20
Reinforcement of network bandwidth Network configuration for the new KEKCC • External link for LHCONE has been extended from 40G to 40Gx2. on Oct. 21st. • IPv4/6 dual stack and Jumbo Frame support are available for the grid instance for non LHCONE connection. • IPv6 for LHCONE is in preparation. S. Suzuki 2021-03-17 Tomoaki Nakamura, KEK-CRC 21
Miscellaneous The CVMFS repository for Belle II: belle.kek.jp • Belle II has originally started with belle.cern.ch • Two replicas (Stratum-1s) in each region • IHEP/KEK in Asia • DESY/RAL in EU • BNL in the US • FNAL as the second site candidate in the US • Many thanks for the support from Dave and Jakob as the CVMFS coordination group • Distributing client setup files Hosting replicas for ATLAS CVMFS repositories • ICEPP/Tokyo-LCG2 is responsible for ATLAS • Avoiding to have two ore more Stratum-1s in the same country Data Management Evolutions • Completed to migrate to Rucio/BNL in January 2021 • LFC will move on to the decommissioning phase and retire in summer 2021 2021-03-17 Tomoaki Nakamura, KEK-CRC 22
Connectivity of International network Feb. 2019: 20 Gbps to 100 Gbps to Amsterdam Mar. 2019: 100 Gbps to New York via Los Angels Mar. 2019: 100 Gbps New York to Amasterdam Sep. 2017: LHCONE for Asian sites at HK by JGN HK https://www.nii.ac.jp/en/news/release/2019/0301.html 2021-03-17 Tomoaki Nakamura, KEK-CRC 23
International network in future SINET6: JFY2022 • to New York and Los Angels: 100Gbps x2 • to Amsterdam: 100Gbps x 2 • to Singapore: 100Gbps • to Guam: 100Gbps HK https://www.nii.ac.jp/en/news/release/2019/0301.html 2021-03-17 Tomoaki Nakamura, KEK-CRC 24
Summary The new KEK central computer system (KEKCC) has been launched on Sep. 1st, 2020. • Basic system functionality is all available not only for the data analysis system but also for the IT infrastructure including Grid-CA, email, Mailing list, Web, Indico, Wiki, Online storage, etc. • A lot of minor issues still remain. • Performance and parameter tuning, and so on. • Need to upgrade RHEL6 based Grid instances, e.g. StoRM, VOMS. • Need to explore the quick DB upgrade of PostgreSQL for Belle II AMGA (Currently Ver. 8.4.20, ~250GB DB takes 1 week to take full dump) • IPv6 advertisement for LHCONE will be available at the next Summer. Computing requirements from the next generation experiments hosted by and related to KEK are becoming high. • Actually, we support Belle II, ILC, KAGRA and many pilot projects. • Several projects have interests in going to utilize the Grid infrastructure. • J-PARC (muon g-2) • T2K / Hyper Kamiokande • LightBIRD • Other small experiments 2021-03-17 Tomoaki Nakamura, KEK-CRC 25
You can also read