PowerPoint Template

1
Particle Nuclear Physics Session on HPC Asia
2007
KREONET
Korea Research Environment Open Network
KREONET Resource Status Strategical Method for
HEP
Wonhyuk Lee
livezone_at_kisti.re.kr
2
Outline

• Network Resource Status on KREONET
• 2. Science Cyber Infrastructure
• 3. Performance Enhancement Activities
• 4. QA and Comments

3

• Network Resource Status on KREONET
• 2. Science Cyber Infrastructure
• 3. Performance Enhancement Activities
• 4. QA and Comments

4
Whats KREONET?

• Korea Research Environment Open NETwork
• National high-performance research network for
  RD community in Korea
• About 300 RD members Universities, government
  industrial institutes, government
  organizations, etc.
• High-capacity access network(1Gbps 10Gbps) 48
  major institutions/universities
• Nation-wide optical gigabit network facility
• 14 GigaPoPs in 13 regions (5Gbps 20Gbps)
• KREONet2 (Next generation advanced network)
• 2 Overseas direct link for international
  collaboration
• Advanced network engineering
• IPv6, QoS, Multicast, Traffic Measurement,
  Security, GHPN, etc.
• Nation-wide 6KREONET and Mbone
• Support advanced application research (Grid,
  e-Science)

5
KREONET/KREONet2
Grid/e-Science based KREONET/KREONet2/SuperSIReN
? joining global Cyber Infrastructure 5-10G
Lambda Network,-1G Access link
High-Performance scientific Facilities (High-Perfo
rmance Cluster/Supercomputers, Storage,
Experimental Facilities, Visualization, Access
Grid, DB Servers, etc.)
6
KREONET Backbone Interconnection
US
10G
5G
5G
10G
20G
5G
10G
5G
5G
10G
5G
10G
HK
14 Giga-Pops in Domestic 2 Giga-Pops in
International
2.5G
7
KREONET Backbone Topology
8
Hybrid Backbone Network
Packet Switched Networks
Optical Circuit Switched Networks
Incheon
Seoul
Seoul
Incheon
WDM 120G
Suwon
Suwon
10?20Gbps
Daejeon
Daejeon
Cheonan
Cheonan
SuperSIReN
Pohang
Pohang
Jeonju
Daegu
Daegu
10?40Gbps
Jeonju
5?10 Gbps
Kwangju
Kwangju
Changwon
Busan
Busan
Changwon
Jeju
Jeju
Designed this facility to provide dedicated
channel for the researchers who want
point-to-point connectivity to transmit their
large sized data with very high quality and low
latency for their applications.
9
GLORIAD(1)

• GLORIAD (GLObal RIng Network for Advanced
  Applications Development)
• Global Ring topology for advanced science
  applications
• Started as the Little GLORIAD, founded by US,
  Russia, China
• 4th core member participation on June. 2004
• Korean government (MOST) decided to fund for
  joining to GLORIAD consortium.
• GLORIAD/IRNC was finally awarded by NSF, Jan.
  2005.
• Now, 11 nations participate in this project
• Essential to support advanced application
  developments
• HEP, ITER, Astronomy, Earth System, BioMedical,
  HDTV etc.
• The BIG GLORIAD 10Gbps Networking
• (Canada)-US- (Netherlands)-Russia-Chi
  na-Korea

10
GLORIAD(2)

• GLORIAD is the first round-the-world
  high-performance networks jointly established by
  China, United States, Russia and Korea with
  10Gbps optical networking tools that improve
  networked collaboration with e-Science and Grid
  applications

GLORIAD (USA-Russia-China-Korea)
Novosibirsk
Amsterdam
10Gbps
Toronto
Moscow
Daejeon
Beijing
Chicago
Seattle
Hong Kong
Russia Kurchatov Inst. MIST
U.S.A UT, ORNL NSF
China CNIC CAS
Korea KISTI MOST
Canada
Netherlands
11
GLORIAD(3)
Russia (Novosibirsk)
CANARIE, Canada
Calgary
Toronto
Moscow
NYC
US
Seattle
EU
10G (KR-US)
Chicago
China (Beijing)
Amsterdam
KISTI, Korea (Daejeon)
CNIC, China (Hong Kong)
Esnet
MREN
SURFnet
StarLight
6TAP
CSTNET
12
Network infra(2)
Stockholm NorthernLight
New York MANLAN
10 Gbit/s

• Multidomain

IEEAF 10 Gbit/s
10 Gbit/s
2.5 Gbit/s
2.5 Gbit/s
10 Gbit/s
GLORIAD 10 Gps
SURFnet 10 Gbit/s
CAnet4
KREONET KRLight
Amsterdam NetherLight
Chicago StarLight
10 Gbit/s
NSF 10 Gbit/s
IEEAF
WIDE
10 Gbit/s
10 Gbit/s
2.5 Gbit/s
SURFnet 10 Gbit/s
Tokyo APAN
2.5 Gbit/s
10 Gbit/s
Prague CzechLight
Geneva CERN
London UKLight
Source Kees Neggers, SURFnet

• 2 National Peers , 12 International Peers RD
  Network

13
Lightpath on KREONET
e-Science Applications
Supercomputing Applications
Other high-end applications
Grid Applications
UCLP(User Controlled LightPaths)
Lightpath Provisioning System based on User
Applications
Layer 3 Traffic Engineering
KREONet2 Routers
Operations Monitoring
KREONet2 Layer 2 / Layer 1 network
dedicated lightpath
14

• Network Resource Status on KREONET
• 2. Science Cyber Infrastructure
• 3. Performance Enhancement Activities
• 4. QA and Comments

15
Background Integrative Science

• Science Integrated, Big, and ..
• New Integrative Science
• Coupling and mediate
• Relationship with IT and full understanding

Source office of Integrative Activities NSF

• Consumer can be provided with big
  infrastructure and can lead.
• Infra accelerate new integrative science
• Consumer begin to consider..
• What is good? Where is near? ? Suitable for
  own research

16
Background integrative Infra
More Shareable, Big domain
Infrastructure bigger and faster and easy to be..
(USA-Russia enable Fast Data Transfer 1.4TeraBytes
of data in 4.5hour. June 4, 2007)
New devices (Sensor, Wireless..), New
applications, Integrative
More Diversity, Knowledge from data
17
Science and CI
Integrative Science
e-Science or e-Research, i-Science..
web services, workflow, security, Middleware
USER
Cyber-infrastructure
Program, Share, Schedule, Allocate, Virtual,
Scale
Networks
Grid-able
Grid-able
Grid-able
HPC
Databases
Instruments
USER
Tele-scope, micro-scope..
Supercomputers Clusters
BT, NT, Climate, SDSS,, storages
18
SI Network User
Todays Science Infra and Network is oriented for
SP, and each communicate
IP address X.X.X.X come and use our resource.
I want to use telescope and supercomputer for my
research.
Science Resource Farm (Provider)

I service only telescope. You may
find Supercomputer an other site.
NETWORK
ST member
I want to join a video-conference What is your
multicast address? What is IP address on your VC
server?
Service Provider (IDC Center)
The network to communicate with each users and
resource
Where is suitable SI for our research?
Instruments
Service Provider (Portal Service)
USER
Join here. We are waiting for you
ST member

• Where is SI? Where is important area? Where is
  the area deploy QoS?
• Could you keep and provide every sites with
  advanced Network service?

19
CI Network User
Resource and data exist on the CI and are
uncoupled from any specific domain or location
Instrument CI
Multimedia Intensive CI
Computational Resource CI
Multicast QoS enable Low jitter
Internet
Cyber Infras Network
NREN
Each site need stable network. LPs were reserved
Have to transfer bulk data More bandwidth
allocated
Other International RD Network
20

• Network Resource Status on KREONET
• 2. Science Cyber Infrastructure
• 3. Performance Enhancement Activities
• 4. QA and Comments

21
Performance Measurement Tools

• Latency/Delay Measurement Tools
• Ping, synack, Traceroute, MTR, Ping Plotter, LFT,
  Smokeping, Pchar, Iperf, BWCTL, ip route show
  cache, one-way delay measurement nodes, OWAMP,
  AMP
• Bandwidth Utilization Measurement Tools
• MRTG, RRDTool, Cricket, Network weathermap
• Path Probe Tools
• Tracepath, traceproto, Path MTU Discovery
• Network Simulation and Benchmarking Tools
• Network Emulation, Netperf, RUDE/CRUDE, TTCP,
  Tweak Tools, NDT
• Traffic Analysis Tools
• TCPDump, Ethereal, Jettop, Cflowd, flow-tools,
• GEANT2 JRA 1/Internet2 E2E piPES, PerfSONAR

22
Identification of performance problem

• Network
• Network equipment component
• Bad NIC, broken wire/cable, cross-talk
• Network line component
• Congestion, Bandwidth
• Multi/Single mode of Fiber
• Network path
• Etc
• Firewalls, Filters, shapers,
• End-Host
• Network Adapter
• TCP Offload Engines(TOEs), Large Send
  Offload(LSO), Interrupt Coalescence, Checksum
  Offload
• Network chipset
• Quality of network driver
• Half/full duplex mode
• File systems and disks(disk access speed)
• Benchmarking tools Bonnie, Iozone, H
• CPU(clock speed), Main Memory, Front Back side
  bus, I/O bus type/capacity(ATA, SCSI, SATA),
• OS (Kernel)

23
Performance Enhancement Summary

• On troubleshooting network performance problems,
  you must identify which part (network or
  end-host) is bottleneck point.
• For LFNs, the default parameters for TCP are too
  conservative.
• When you make and use the program (application),
  related to network, you need to consider
  parameters for send and receive buffer size.
• Tuning congestion window size and using
  multi-stream is a practical method for
  performance enhancement.
• Most WAN backbone links are at least 1Gbps, so
  the bottleneck link are likely to be local
  networks.
• Lightpath allocation is one of the best solutions
  for network performance problems, but the
  performance (access/read/write) of hosts file
  system is still bottleneck point.

24
Phoebus

• Its a project that started by Internet2
• And now we are doing this project on KREONET
• Phoebus aims to narrow the performance gap by
  bringing revolutionary networks like HOPI to
  users
• Phoebus is another name for the mythical Apollo
  in his role as the sun god
• Phoebus stresses the session concept to enable
  multiple network/transport infrastructures to be
  catenated
• Phoebus builds on an earlier project called the
  Logistical Session Layer (LSL)

25

• Network Resource Status on KREONET
• 2. Science Cyber Infrastructure
• 3. Performance Enhancement Activities
• 4. QA and Comments

26
Q A Comments?
27
Your Company Slogan
Thank You.
livezone_at_kisti.re.kr