Transcript GLIF: Linking the Globe with LIGHT

Aloha Kakahiaka !
GLIF
Linking the Globe with
LIGHT
Gigi Karmous-Edwards
Principal Scientist
MCNC
[email protected]
APAN 2008, Hawaii
Agenda
1.
2.
3.
4.
5.
6.
7.
What is GLIF?
Why does GLIF exists?
How Does GLIF function?
What has GLIF accomplished?
Virtualization
The Many Challenges ahead
Conclusions
What is GLIF?
Global Lambda Integrated Facility
www.glif.is
•
GLIF is an international virtual organization
that promotes the paradigm of lambda
networking
•
GLIF participants jointly make lambdas
available as an integrated global facility for
use in data-intensive research
•
GLIF brings together leading networking
engineers and researchers worldwide, who
collaborate to identify and solve challenges
for a Global facility
What is GLIF?
•
GLIF is an international virtual organization
managed as a cooperative activity with
‘participants’ rather than ‘members’ with a
lightweight governance structure.
•
Open to anybody sharing the vision of optical
interconnection of different facilities, who
voluntarily contributes network resources
(e.g. equipment, lambdas) or actively
participates in relevant activities.
•
Please join the mailing list if you have an
interest in being part of the solution for
facilitating global lambda networks for
research and education.
GLIF …. More resources are now available, next version in two weeks!
Why GLIF exists? … E-science
• Researchers need to do their work globally
• E-science: global, large scale scientific collaborations
enabled through distributed computational and
communication infrastructure
• Combines scientific instruments and sensors,
distributed data archives, computing resources and
visualization to solve complex scientific problems
• In physics, molecular biology, environmental, Health,
Entertainment, etc.
• Future - this facility will be useful for K-20 education
not just E-Scientist
Developing a Global E-science Laboratory (GEL)
• Korea’s HVEM
• One of a kind in the world
• Provide global access to unique
instruments for the purpose of
advancing science for humanity
• WEB service interface
• High capacity optical network for
output
•
•
•
•
Viewing the real-time video from the CCD camera
Accessing or manipulating the 2-D or 3-D images
Generating the workflow specification and requesting the
workflow to be executed
Searching the images or video files, papers, and
experiments in the databases or storages
Hyuck Han, Hyungsoo Jung, Heon Y. Yeom, Hee S. Kweon, and Jysoo Lee
”HVEM Grid: Experiences in Constructing an Electron Microscopy Grid”
Accommodating Researchers
•
•
Need high Capacity - 1Gbs - 10Gbs or more
Need QoS - difficult to guarantee w/ routed
network
•
Cannot disrupt current users with their large
flows
So… We need Hybrid Networking (IP + lambda
networking)
•
•
Lightpath: high quality and high capacity
optical end-to-end network connection
Lightpaths provide applications with dedicated
bandwidth with fixed characteristics at
relatively low costs and with added security
The GLIF Story …
•
September 2001: first Lambda Workshop in
Amsterdam followed by open Lambda
Workshop organized by TERENA
•
Second Lambda Workshop in 2002 in
Amsterdam was attached to iGrid2002, hosted
by Science Park Amsterdam
•
August 2003: third Lambda Workshop in
Reykjavik hosted by NORDUnet and attached
to the NORDUnet 2003 Conference -GLIF
name created
www.GLIF.is
How GLIF functions?
There are Four working groups:
– Governance
– Research and Applications
– Technical
– Control Plan
•Secretariat functions by TERENA
•Holds Annual meeting
– Next Meeting - 8th Annual Global LambdaGrid
Workshop, Seattle, USA, 1-2 October 2008
•Tech and Control working groups also hold semiannual meetings (past weekend)
GLIF Working Groups
Governance and Growth (GOV) Working Group
Chair: Kees Neggers (SURFnet)
Goals: To identify future goals in terms of
lambdas, connections and applications
support, and to decide what cross-domain
policies need to be put in place.
Research and Applications (RAP) Working Group
Chair: Maxine Brown (UIC) & Larry Smarr (UCSD)
Goals: To train a new generation of scientists on
the use of super-networks.
GLIF Working Groups
Technical Issues (Tech) Working Group
Co-Chairs: Erik-Jan Bos (SURFnet) & René Hatem
(CANARIE)
Goals: To design and implement an international
LambdaGrid infrastructure, identify which
equipment is being used, what connection
requirements are required, and which functions
and services should be provided.
Control Plane and Grid Integration Middleware
Working Group
Chair: Gigi Karmous-Edwards (MCNC)
Goals: To agree on the interfaces and protocols
that talk to each other on the control planes of the
contributed Lambda resources.
GLIF RAP working group Accomplishments
•
•
•
•
•
•
•
•
Documented enabling technologies (middleware, control plane
software) and what applications they enable (e.g., DRAGON, UCLP,
etc)
Documented countries’ activities (feedback to NRENs)
Helped applications get started
Provides a resource for groups trying to get funding for GLIFrelated activities; GLIF “branding” adds credibility
Document applications (brief descriptions with URL pointers) (I
will create template and forward to RAP email list)
Developed a GLIF primer (how to find, educate, promote
applications)
Provided PR: What can GLIF do for you?
Provided PR: Promote domain-specific applications (eVLBI,
CineGrid, etc) (provide inspiration and motivation to potential new
applications within countries)
GLIF Tech working group Accomplishments
Chairs: Erik-Jan Bos and Rene Hatem, Secretary: Kevin
Meynell
•
Developed concept of GOLEs
•
Documented in a centralized database all technical
information on contributed resources
•
Developed best practices and issues document for
Hybrid Networking
•
Developed best practices document for fault
resolutions
•
Hold monthly resource update calls
•
Share Open source toolkits such as TL1 toolkit
•
And more…
GOLES
GLIF Open Lightpath Exchanges
•
GLIF lambdas are interconnected
through established lightpath
exchange points known as GOLEs
•
GOLEs are comprised of equipment
capable of terminating lambdas and
performing lightpath switching,
allowing end-to-end connections
•
GOLEs have an open connection policy
GOLES,
example of a GOLE, NetherLight
Current GLIF Resources
•
•
•
•
•
•
•
•
•
•
•
AMPATH - Miami
CERN - Geneva
•
CzechLight - Prague
HKOEP - Hong Kong
KRLight - Daejoen
MAN LAN - New York
•
MoscowLight - Moscow •
NetherLight - Amsterdam
•
NGIX-East - Washington•
DC
NorthernLight •
Copenhagen
•
Pacific Wave (Los
Angeles)
PacificWave (Seattle)
PacificWave(Sunnyva
le)
StarLight - Chicago
T-LEX - Tokyo
TaiwanLight - Taipei
UKLight - London
AARNet,
US LHCNet
GLIF Control Plane and Grid
Middleware Integration wg
Chair: Gigi Karmous-Edwards, Sectretary: Licia Florio
•
Virtualization of Networking resources as well as other key
resources (compute, storage, instruments, etc) via “ondemand” and “advanced reservations”
Agreed to adopt Network Description Language (NDL) based
on RDF
Work closely with two OGF working groups for
standardization
•
•
–
–
•
•
•
•
•
Grid High Performance Networking wg
Network Markup Language wg
Shared current research experiments and open source code
for controlling lightpaths
Developed an architecture for next generation lambda
resources coordinated with other key resources
Agreed to focus on Generic Network Interface (GNI)
Comparing existing APIs similar to GNI
Will have an initial GNI specification by October meeting
KISS
Keep it Simple
and
Smart!
----------------
Akamai
User
GLIF Grid
Resource
Registry
GAI
Resource
Registry
RB-A
RB-B
Resource
Registry
GNI
NRM-A
NRM-B
CRM-A
Network-B
Network-A
CRM-B
IRM-B
SRM-A
Grid Administrative Domain - B
Grid Administrative Domain - A
RB: Resource Broker
DNRM: Domain Network Resource Manager
CRM: Compute Resource Manager
IRM: Instrument Resource Manager
SRM: Storage Resource Manager
GAI: Grid Application Interface
GNI: Grid Network Interface
GCI: Grid Compute Interface
GSI: Grid Storage Interface
GII: Grid Instrument Interface
Publish Resource Information
Publish/Subscribe Broker + Resource
Information / References
GAI
Security/AAA
Publish/
Subscribe
RB
GLIF Grid
Resource
Registry
Policy Engine
Resource
Registry
Resource
Meta-scheduler
Request
Processor
Multi-domain
Path Computation
Monitoring
Discovery
- Fault Mgmt
- Performance
Resource Co-allocation
Static
Information
(Policy, etc)
HARC Acceptors
RMs RMs
HARC
GNI
GNI GSI, GII, GxI, etc
GNI
Publish Information
to ERB
NRM
Security/AAA
Resource
Repository
Policy Engine
Path
Computation
Publish/
Subscribe
Request
Processor
Network
Management:
- Fault Mgmt
- Performance
Reservation
timetable
Monitoring
Discovery
Static
Information
(Policy, etc)
Topology/
Discovery
Resource
Allocation
e.g. TL1,
SNMP,
XML,MDS,
etc.
e.g. TL1, SNMP
Domain A
NCSU’s Virtual Computing Lab
(VCL) vcl.ncsu.edu
• MCNC experimenting with new Virtual Compute
Services for NC’s K-20 community
• Reservation and Provisioning system
– Allocates nodes to users on a reservation basis
– Can be now (on-demand) or future (schedule in
advance)
– Can allocate both single nodes and clusters of
nodes
– Reservation lengths are policy driven
• selection of 1-4 hours
• Or open end time allow a month or more
Virtual Computing Lab
• Will host for NCSU 1000 nodes at MCNC this year
• Pilots are under way with K-20 type users
• IBM BladeCenter Blade Servers
• Housed in a datacenter - IBM’s Energy efficiency
doors
• Standalone workstations
• Housed anywhere; we include our lab machines
when the labs are closed
• Working on Sun Blade servers
– VCI partners are working Dell and HP blades
– Can easily be moved between HPC cluster and VCL
system
– We move nodes to HPC during student breaks
Differentiator: User to Image to Resource Mapping, Management & Provenance
“Images”
Users
“Application” Image Stack
End-User
Access
…
RDP,
VNC,
…
Client
X-Win
Apps.
Work
Vis
Flow
Services Other …
Services
e.g.,
Web
Sphere
Apps
Middleware
e.g. LSF
Image
OS:
Win
e.g.,
Web
Sphere
Apps
OS:
VCL
Manager
Virtual
Layer
Linux
e.g.,
VMWare,
XEN,
MSVS2500,..
Win
Linux
Other …
Hardware
H/W Resources
Blades, servers, desktops, storage
Undifferentiated Local or distributed
xCAT
VCL code IBM TM
WebServer DataBase
Etc.
Simplicity, Flexibility, Reliability
Scalability, Economy
…
Some Stats
• About 1000 blades (cca 140 used for VCL individual
seats, the rest for VCL HPC cycles), plus several
hundred idle student laboratory machines.
• Environment base-lines are typically Windows and
Linux with a variety of applications. Depending on
how demanding an application is, service may be
virtualized (VMWare) or bare-metal.
• About 70,000 single-seat image reservations per
semester. Fall 2007, peaked at about 2,500
reservations per day.
• Serving population of 30,000 students (in a
semester there may be about 6,000 unique users).
• Most of the “individual seat” requests are ondemand “Now” reservations: cca 90% of requests
• System availability: about 99%
Issues and Challenges
•
•
•
•
•
•
•
•
•
•
Key Challenges with Hybrid networking - effect on
IP while having dynamic lambdas
Coordination of network resources and other Grid
resources
Two phase commit for all involved resources - KISS
Topology Abstractions - including end points - or
services
Monitoring - MonALISA, PerfSONAR….
Advertising resources globally - agree on what and
how to represent resources… NDL etc.
Policy
Different implementations of each component (no
need to standardize on how things are done - just
interfaces)
Agree on Functional components
Focus on a couple of KEY interfaces (low set of
options - use lowest common denominator)
Prioritize - GNI …
Conclusions
Conclusions
•
A Global Integrated Facility is necessary for the support of
both Scientific Research, Education, and networking research.
Everyday there are more requests for use and more resources
contributed.
•
GLIF currently behaves as a Global collaborative testbed
•
Our goal is to provide Global virtualization of shared resources
, including network lambdas, compute, storage, instruments,
etc.
•
Next Generation Networks will be a hybrid of of routed and
lambda switched networks. (not just for high-end research)
•
The Research networks (NRENs and Gov sponsored testbeds)
are taking these bold steps on GLIF, testbed infrastructures…
apply lessons learned to production quickly.
•
International Collaboration is a very Key ingredient for the
future of Scientific discovery and education - The Optical
network plays the most critical role in achieving this!
Mahalo
Gigi Karmous-Edwards
[email protected]
APAN 2008