Transcript title
HENP Grids and Networks Global Virtual Organizations Harvey B Newman
FAST Meeting, Caltech July 1, 2002
http://l3www.cern.ch/~newman/HENPGridsNets_FAST070202.ppt
Computing Challenges: Petabyes, Petaflops, Global VOs
Geographical dispersion: Complexity: Scale: of people and resources the detector and the LHC environment Tens of Petabytes per year of data 5000+ Physicists 250+ Institutes 60+ Countries Major challenges associated with: Communication and collaboration at a distance Managing globally distributed computing & data resources Cooperative software development and physics analysis New Forms of Distributed Systems: Data Grids
Four LHC Experiments: The Petabyte to Exabyte Challenge
ATLAS, CMS, ALICE, LHCB Higgs + New particles; Quark-Gluon Plasma; CP Violation
Data stored CPU ~40 Petabytes/Year and UP; 0.30 Petaflops and UP 0.1 to 1 Exabyte (1 EB = 10 18 Bytes) (2007) (~2012 ?) for the LHC Experiments
LHC: Higgs Decay into 4 muons (Tracker only); 1000X LEP Data Rate (+30 minimum bias events) Reconstructed tracks with pt > 25 GeV All charged tracks with pt > 2 GeV 10 9 events/sec, selectivity: 1 in 10 13 (1 person in a thousand world populations)
LHC Data Grid Hierarchy
~PByte/sec CERN/Outside Resource Ratio ~1:2 Tier0/(
Tier1)/(
Tier2) ~1:1:1 Experiment Online System ~100-400 MBytes/sec
Tier 0 +1
CERN 700k SI95 ~1 PB Disk; Tape Robot
Tier 1
IN2P3 Center ~2.5-10 Gbps RAL Center INFN Center FNAL: 200k SI95; 600 TB 2.5-10 Gbps
Tier 3
~2.5-10 Gbps
Tier 2
Tier2 Center
Institute Institute
Physics data cache Workstations 0.1–10 Gbps
Tier 4
Physicists work on analysis “channels” Each institute has ~10 physicists working on one or more channels
Emerging
Data Grid
User Communities
Grid Physics Network (GriPhyN)
ATLAS, CMS, LIGO, SDSS
Particle Physics Data Grid (PPDG) Int’l Virtual Data Grid Lab (iVDGL)
NSF Network for Earthquake Engineering Simulation (NEES)
Integrated instrumentation, collaboration, simulation Access Grid; VRVS: supporting group-based collaboration
And
Genomics, Proteomics, ...
The Earth System Grid and EOSDIS Federating Brain Data Computed MicroTomography … Virtual Observatories
Projects
PPDG I
GriPhyN
EU DataGrid EU
PPDG II (CP) USA
iVDGL
USA USA
USA
DataTAG
GridPP
HENP Related Data Grid Projects
EU UK
DOE NSF $11.9M + $1.6M 2000-2005 EC €10M 2001-2004
DOE NSF EC
$2M
$9.5M
$13.7M + $2M €4M PPARC >$15M
1999-2001
2001-2004 2001-2006 2002-2004 2001-2004
LCG (Ph1) CERN MS 30 MCHF 2002-2004
Many Other Projects of interest to HENP
Initiatives in US, UK, Italy, France, NL, Germany, Japan, …
Networking initiatives: DataTAG, AMPATH, CALREN-
XD…
US Distributed Terascale Facility: ($53M, 12 TeraFlops, 40 Gb/s network)
Daily, Weekly, Monthly and Yearly Statistics on 155 Mbps US-CERN Link
20 - 100 Mbps Used Routinely in ’01 BaBar: 600 Mbps Throughput in ‘02 BW Upgrades Quickly Followed by Upgraded Production Use
Tier A
"Physicists have indeed foreseen to test the GRID principles starting first from the Computing Centres in Lyon and Stanford (California). A first step towards the ubiquity of the GRID." Pierre Le Hir
Le Monde 12 april 2001 3/2002 D. Linglin: LCG Wkshop
Two centers are trying to work as one: -Data not duplicated -Internationalization transparent access, etc…
CERN-US Line + Abilene Renater + ESnet
RNP Brazil (to 20 Mbps) FIU Miami/So. America (to 80 Mbps)
Transatlantic Net WG (HN, L. Price) Bandwidth Requirements [*]
CMS 2001 2002 2003 2004 2005 2006 100 200 300 600 800 2500 ATLAS BaBar 50 100 300 600 800 2500 300 600 1100 1600 2300 3000 CDF D0 BTeV DESY 100 300 400 2000 3000 6000 400 1600 2400 3200 6400 8000 20 40 100 180 100 210 200 240 300 270 500 300 See http://gate.hep.anl.gov/lprice/TAN CERN BW [*] 155 310 622 2500 5000 10000 20000 Installed BW. Maximum Link Occupancy 50% Assumed
MONARC: CMS Analysis Process Hierarchy of Processes (Experiment, Analysis Groups,Individuals) RAW Data 3000 SI95sec/event 1 job year 3000 SI95sec/event 3 jobs per year Experiment Wide Activity (10 9 events) Reconstruction Re-processing 3 Times per year New detector calibrations Or understanding Monte Carlo 5000 SI95sec/event 25 SI95sec/event ~20 jobs per month ~20 Groups’ Activity (10 9
10 7 events) Selection Iterative selection Once per month Trigger based and Physics based refinements 10 SI95sec/event ~500 jobs per day ~25 Individual per Group Activity (10 6 –10 7 events) Analysis Different Physics cuts & MC comparison ~Once per day Algorithms applied to data to get results
Tier0-Tier1 Link Requirements Estimate: for Hoffmann Report 2001
1) Tier1
Tier0 Data Flow for Analysis 2) Tier2
Tier0 Data Flow for Analysis 3) Interactive Collaborative Sessions (30 Peak) 4) Remote Interactive Sessions (30 Flows Peak) 5) Individual (Tier3 or Tier4) data transfers Limit to 10 Flows of 5 Mbytes/sec each TOTAL Per Tier0 - Tier1 Link 0.5 - 1.0 Gbps 0.2 - 0.5 Gbps 0.1 - 0.3 Gbps 0.1 - 0.2 Gbps 0.8 Gbps 1.7 - 2.8 Gbps NOTE:
Adopted by the LHC Experiments; given in the Steering Committee Report on LHC Computing: “1.5 - 3 Gbps per experiment”
Corresponds to ~10 Gbps Baseline BW Installed on US-CERN Link
Report also discussed the effects of higher bandwidths
For example all-optical 10 Gbps Ethernet + WAN by 2002-3
Tier0-Tier1 BW Requirements Estimate: for Hoffmann Report 2001
Does Not Include more recent ATLAS Data Estimates
270 Hz at 10 33
400 Hz at 10 34 Instead of 100Hz Instead of 100Hz
2 MB/Event Instead of 1 MB/Event ?
Does Not Allow Fast Download to Tier3+4 of “Small” Object Collections
Example: Download 10 7 Events of AODs (10 4 Bytes)
100 Gbytes; At 5 Mbytes/sec per person (above) that’s 6 Hours !
This is a still a rough, bottoms-up, static, and hence Conservative Model.
A Dynamic distributed DB or “Grid” system with Caching, Co-scheduling, and Pre-Emptive data movement may well require greater bandwidth
Does Not Include “Virtual Data” operations; Derived Data Copies; Data-description overheads
Further MONARC Model Studies are Needed
Maximum Throughput on Transatlantic Links (155 Mbps) *
8/10/01 105 Mbps reached with 30 Streams: SLAC-IN2P3 9/1/01 102 Mbps in One Stream: CIT-CERN 11/5/01 125 Mbps in One Stream (modified kernel): CIT-CERN 1/09/02 190 Mbps for One stream shared on 2 155 Mbps links 3/11/02 120 Mbps Disk-to-Disk with One Stream on 155 Mbps link (Chicago-CERN) 5/20/02 450 Mbps SLAC-Manchester on OC12 with ~100 Streams 6/1/02 290 Mbps Chicago-CERN One Stream on OC12 (mod. Kernel) Also see http://www-iepm.slac.stanford.edu/monitoring/bulk/; and the Internet2 E2E Initiative: http://www.internet2.edu/e2e
Some Recent Events: Reported 6/1/02 to ICFA/SCIC
Progress in High Throughput: 0.1 to 1 Gbps
Land Speed Record: SURFNet – Alaska (IPv6)
(0.4+ Gbps) SLAC – Manchester (Les C. and Richard H-J) (0.4+ Gbps)
Tsunami (Indiana) (0.8 Gbps UDP) Tokyo – KEK (0.5 – 0.9 Gbps)
Progress in Pre-Production and Production Networking
10 Mbytes/sec FNAL-CERN (Michael Ernst)
15 Mbytes/sec disk-to-disk Chicago-CERN (Sylvain Ravot)
KPNQwest files for Chapter 11; Stops network yesterday.
Near Term Pricing of Competitor (DT) ok.
Unknown impact on prices and future planning in the medium and longer term
Transoceanic Networking Integrated with the Abilene, TeraGrid, Regional Nets and Continental Network Infrastructures in US, Europe, Asia, South America Baseline BW for the US-CERN Link: HENP Transatlantic WG (DOE+NSF ) Link Bandwidth (Mbps) 20000 15000 10000 5000 0 FY2001 FY2002 FY2003 FY2004 FY2005 FY2006 BW (Mbps) 310 Baseline evolution typical of major HENP links 2001-2006 622 2500 5000 10000 20000
US-CERN Link: 622 Mbps this month DataTAG 2.5 Gbps Research Link in Summer 2002 10 Gbps Research Link by Approx. Mid-2003
Total U.S. Internet Traffic
100 Pbps 10 Pbps 1 Pbps 100Tbps 10Tbps 1Tbps 100Gbps 10Gbps 1Gbps 100Mbps 10Mbps 1Mbps 100Kbps 10Kbps 1Kbps 100 bps 10 bps 1970 1975 1980 96 1985 Limit of same % GDP as New Measurements Voice Crossover: August 2000 ARPA & NSF Data to 1990 Voice 2.8X/Year 1995 2000 Projected at 3/Year 4X/Year 2005 2010 U.S. Internet Traffic Source: Roberts et al., 2001
Internet Growth Rate Fluctuates Over Time
U.S. Internet Edge Traffic Growth Rate
6 Month Lagging Measure 4.50
4.00
10/00 –4/01 Growth Reported 3.6/year 10/00 –4/01 Growth Reported 4.0/year 3.50
3.00
2.50
2.00
1.50
1.00
0.50
Average: 3.0/year 0.00
Jan 00 Apr 00 Jul 00 Oct 00 Jan 01 Apr 01 Jul 01 Oct 01 Jan 02
Source: Roberts et al., 2002
AMS-IX Internet Exchange Throughput Accelerating Growth in Europe (NL) Monthly Traffic 2X Growth from 8/00 - 3/01; 2X Growth from 8/01 - 12/01
↓
6.0 Gbps Hourly Traffic 3/22/02 4.0 Gbps 2.0 Gbps
ICFA SCIC Meeting March 9 at CERN: Updates from Members
Abilene Upgrade
from 2.5 to 10 Gbps Additional scheduled lambdas planned for targeted for targeted applications: Pacific and National Light Rail
US-CERN
Upgrade On Track: to 622 Mbps in July; Setup and Testing Done in STARLIGHT
2.5G Research Lambda by this Summer: STARLIGHT-CERN 2.5G Triangle between STARLIGHT (US), SURFNet (NL), CERN
SLAC + IN2P3 (BaBar)
Getting 100 Mbps over 155 Mbps CERN-US Link
50 Mbps Over RENATER 155 Mbps Link, Limited by ESnet 600 Mbps Throughput is BaBar Target for this Year
FNAL
Expect ESnet Upgrade to 622 Mbps this Month
Plans for dark fiber to STARLIGHT underway, could be done in ~4 Months; Railway or Electric Co. provider
ICFA SCIC: A&R Backbone and International Link Progress
GEANT Pan-European Backbone
( http://www.dante.net/geant Now interconnects 31 countries Includes many trunks at 2.5 and 10 Gbps )
UK
2.5 Gbps NY-London, with 622 Mbps to ESnet and Abilene
SuperSINET (Japan): 10 Gbps IP and 10 Gbps Wavelength
Upgrade to Two 0.6 Gbps Links, to Chicago and Seattle Plan upgrade to 2 X 2.5 Gbps Connection to US West Coast by 2003
CA*net4 (Canada):
Interconnect customer-owned dark fiber nets across Canada at 10 Gbps, starting July 2002 “Lambda-Grids” by ~2004-5
GWIN (Germany): Connection to Abilene Upgraded to 2 X 2.5 Gbps early in 2002
Russia
Start 10 Mbps link to CERN and ~90 Mbps to US Now
2.5
10 Gbps Backbone 210 Primary Participants All 50 States, D.C. and Puerto Rico 80 Partner Corporations and Non-Profits 22 State Research and Education Nets 15 “GigaPoPs” Support 70% of Members Caltech Connection with GbE to New Backbone
National R&E Network Example Germany: DFN TransAtlanticConnectivity Q1 2002
2 X OC12 Now: NY-Hamburg and NY-Frankfurt
ESNet peering at 34 Mbps
Upgrade to 2 X OC48 expected in Q1 2002
Direct Peering to Abilene and
Canarie expected UCAID will add another 2 OC48’s; Proposing a Global Terabit Research Network (GTRN)
STM 16
FSU Connections via satellite: Yerevan, Minsk, Almaty, Baikal
Speeds of 32 - 512 kbps
SILK Project (2002): NATO funding
Links to Caucasus and Central Asia (8 Countries)
Currently 64-512 kbps
Propose VSAT for 10-50 X BW: NATO + State Funding
National Research Networks in Japan
NIFS
SuperSINET
Started operation January 4, 2002
Support for 5 important areas:
Nagoya U
HEP,
Space/Astronomy, GRIDs Provides 10
’s:
Genetics, Nano-Technology, 10 Gbps IP connection Nagoya
7 Direct intersite GbE links Osaka
Some connections to 10 GbE in JFY2002
Osaka U
HEPnet-J
Will be re-constructed with MPLS-VPN in SuperSINET
Kyoto U ICR Kyoto-U
Proposal: Two TransPacific 2.5 Gbps Wavelengths, and Japan-CERN Grid Testbed by ~2003
Internet
NIG
Tokyo
ISAS NAO
IP
WDM path
IP router
KEK NII Chiba NII Hitot.
U Tokyo IMS U-Tokyo
DataTAG Project
NewYork UK SuperJANET4 It GARR-B STARLIGHT GENEVA GEANT ABILEN E ESNET NL SURFnet STAR-TAP CALRE N Fr Renater
EU-Solicited Project. CERN , PPARC (UK), Amsterdam (NL), and INFN (IT); and US (DOE/NSF: UIC, NWU and Caltech) partners
Main Aims:
Ensure maximum interoperability between US and EU Grid Projects Transatlantic Testbed for advanced network research
2.5 Gbps Wavelength Triangle 7/02 (10 Gbps Triangle in 2003)
TeraGrid (www.teragrid.org) NCSA, ANL, SDSC, Caltech
A Preview of the Grid Hierarchy and Networks of the LHC Era Abilene Chicago Indianapolis Urbana Caltech San Diego OC-48 (2.5 Gb/s, Abilene) Multiple 10 GbE (Qwest) Multiple 10 GbE (I-WIRE Dark Fiber)
I-WIRE
UIC ANL Starlight / NW Univ
Multiple Carrier Hubs
Ill Inst of Tech Univ of Chicago NCSA/UIUC Indianapolis (Abilene NOC) Idea to extend the TeraGrid to CERN Source: Charlie Catlett, Argonne
CA ONI, CALREN-XD + Pacific Light Rail Backbones (Proposed)
Also: LA-Caltech Metro Fiber; National Light Rail
Key Network Issues & Challenges
Net Infrastructure Requirements for High Throughput
Packet Loss must be ~Zero (at and below 10 -6 )
I.e. No “Commodity” networks
Need to track down uncongested packet loss
No Local infrastructure bottlenecks
Multiple Gigabit Ethernet “clear paths” between
selected host pairs are needed now To 10 Gbps Ethernet paths by 2003 or 2004
TCP/IP stack configuration and tuning Absolutely Required
Large Windows; Possibly Multiple Streams
New Concepts of Fair Use Must then be Developed
Careful Router, Server, Client, Interface configuration
Sufficient CPU, I/O and NIC throughput sufficient
End-to-end
monitoring and tracking of performance Close collaboration with local and “regional” network staffs
TCP Does Not Scale to the 1-10 Gbps Range
A Short List: Revolutions in Information Technology (2002-7)
Managed
Global Data Grids (As Above) Scalable Data-Intensive Metro and Long Haul
Network Technologies
DWDM: 10 Gbps then 40 Gbps per
; 1 to 10 Terabits/sec per fiber
10 Gigabit Ethernet (See www.10gea.org) 10GbE / 10 Gbps LAN/WAN integration
Metro Buildout and Optical Cross Connects Dynamic Provisioning
Dynamic Path Building
“Lambda Grids” Defeating the “Last Mile” Problem (Wireless; or Ethernet in the First Mile)
3G and 4G Wireless Broadband (from ca. 2003); and/or Fixed Wireless “Hotspots”
Fiber to the Home
Community-Owned Networks
A Short List: Coming Revolutions in Information Technology
Storage Virtualization
Grid-enabled Storage Resource Middleware (SRM)
iSCSI (Internet Small Computer Storage Interface); Integrated with 10 GbE
Global File Systems
Internet Information Software Technologies
Global Information “Broadcast” Architecture
E.g the Multipoint Information Distribution Protocol ([email protected])
Programmable Coordinated Agent Architectures
E.g. Mobile Agent Reactive Spaces (MARS)
by Cabri et al., University of Modena The “Data Grid” - Human Interface
Interactive monitoring and control of Grid resources
By authorized groups and individuals
By Autonomous Agents
Year 2001 2002 2003 2005 2007 2008 2010 2012
HENP Major Links: Bandwidth Roadmap (Scenario) in Gbps
Production
0.155 0.622 2.5 10 2-4 X 10 ~10 X 10 or 1-2 X 40 ~5 X 40 or ~20 X 10 ~Terabit
Experimental
0.622-2.5 2.5 10 2-4 X 10 ~10 X 10; 40 Gbps ~5 X 40 or ~20-50 X 10 ~25 X 40 or ~100 X 10 ~MultiTerabit
Remarks
SONET/SDH SONET/SDH DWDM; GigE Integ. DWDM; 1 + 10 GigE Integration
Switch;
Provisioning 1 st Gen.
Grids 40 Gbps
Switching 2 nd Gen
Grids Terabit Networks ~Fill One Fiber or Use a Few Fibers
One Long Range Scenario (Ca. 2008 12) HENP As a Driver of Optical Networks Petascale Grids with TB Transactions
Problem: Extract “Small” Data Subsets of 1 to 100 Terabytes from 1 to 1000 Petabyte Data Stores Survivability of the HENP Global Grid System, with hundreds of such transactions per day (circa 2007) requires that each transaction be completed in a relatively short time. Example: Take 800 secs to complete the transaction. Then Transaction Size (TB) Net Throughput (Gbps) 1 10 10 100 100 1000 (Capacity of Fiber Today)
Summary: Providing Switching of 10 Gbps wavelengths within ~3 years; and Terabit Switching within 5-10 years would enable “Petascale Grids with Terabyte transactions”, as required to fully realize the discovery potential of major HENP programs, as well as other data-intensive fields.
Internet2 HENP WG [*]
Mission: To help ensure that the required
National and international network infrastructures (end-to-end)
Standardized tools and facilities for high performance and end-to-end monitoring and tracking, and
Collaborative systems
are developed and deployed in a timely manner, and used effectively to meet the needs of the US LHC and other major HENP Programs, as well as the at-large scientific community.
To carry out these developments in a way that is broadly applicable across many fields
Formed an Internet2 WG as a suitable framework: Oct. 26 2001
[*] Co-Chairs: S. McKee (Michigan), H. Newman (Caltech); Sec’y J. Williams (Indiana)
Website: http://www.internet2.edu/henp ; also see the Internet2 End-to-end Initiative: http://www.internet2.edu/e2e
True End to End Experience
User perception
Application
Operating system
Host IP stack
Host network card
Local Area Network
Campus backbone network
Campus link to regional network/GigaPoP
GigaPoP link to Internet2 national backbones
International connections
EYEBALL APPLICATION STACK JACK NETWORK . . .
. . .
. . .
. . .
HENP Scenario Limitations: Technologies and Costs
Router Technology and Costs (Ports and Backplane)
Computer CPU, Disk and I/O Channel Speeds to Send and Receive Data
Link Costs: Unless Dark Fiber (?)
MultiGigabit Transmission Protocols
End-to-End “100 GbE” Ethernet (or something else) by ~2006: for LANs to match WAN speeds
Throughput quality improvements: BW
TCP
< MSS/(RTT*sqrt(loss)) [*]
80% Improvement/Year
Factor of 10 In 4 Years Eastern Europe Far Behind China Improves But Far Behind [*] See “Macroscopic Behavior of the TCP Congestion Avoidance Algorithm,” Matthis, Semke, Mahdavi, Ott, Computer Communication Review 27(3), 7/1997
11900 Hosts; 6620 Registered Users in 61 Countries 43 (7 I2) Reflectors Annual Growth 2 to 3X
Networks, Grids and HENP
Next generation 10 Gbps network backbones are almost here: in the US, Europe and Japan
First stages arriving, starting now
Major transoceanic links at 2.5 - 10 Gbps in 2002-3
Network improvements are especially needed in Southeast Europe, So. America; and some other regions:
Romania, Brazil; India, Pakistan, China; Africa
Removing regional, last mile bottlenecks and compromises in network quality are now
All on the critical path
Getting high (reliable; Grid) application performance across networks means!
End-to-end monitoring; a coherent approach
Getting high performance (TCP) toolkits in users’ hands
Working in concert with AMPATH, Internet E2E, I2 HENP WG, DataTAG; the Grid projects and the GGF