Grids and Peer-to-Peer Networks
for e-Science
PTLIU Laboratory for Community Grids
Geoffrey Fox and Community Grid Staff and Students Computer Science, Informatics, Physics
Indiana University, Bloomington IN 4740
http://grids.ucs.indiana.edu/ptliupages
Summary
n
Grid:
Global Computing Infrastructure with a myriad
of heterogeneous devices connected by diverse
networks
• Measure and study their performance
• Related to but different from classical parallel computing
performance studies
n
Web services:
New object models providing
universality in a service model of electronic capability
• Simulate, data access/storage etc.
• Nodes of application level systems one can model
n
Systems involve multiple devices connected together –
synchronization of these is performance driver
• Communities or virtual organizations are e-Science collective systems
Trends of Importance
n Resources of increasing performance or functionality
• Computers (ASCI, Earth Simulator to TeraGrid), storage,
sensors, networks, PDA’s
• More and more data distributed around the world
n Applications of increasing sophistication
• Size, multi-scales, multi-disciplines
• Compose simulations from different disciplines
n New algorithms and mathematical techniques n Traditional Computer science
• Compilers, Parallelism, Objects, Components
n Grid and Internet Concepts and Technologies
• Enabling new applications
Projected Top 500 Until Year 2009
n First, Tenth, 100th, 500th, SUM of all 500 Projected in Time
Earth Simulator from Japan
http://geofem.tokyo.rist.or.jp/
PACI 13.6 TF Linux TeraGrid
32 32 5 32 32 5 HPS S HPS S ESnet HSCC MREN/Abilene Starlight 10 GbE NCSA 500 Nodes 8 TF, 4 TB Memory240 TB disk
SDSC
256 Nodes 4.1 TF, 2 TB Memory
225 TB disk
Caltech
32 Nodes 0.5 TF 0.4 TB Memory 86 TB disk
Argonne
64 Nodes 1 TF 0.25 TB Memory 25 TB disk
4 Juniper M160 OC-12 OC-48 OC-12 574p IA-32 Chiba City 128p Origin HR Display & VR Facilities
= 32x 1GbE
= 64x Myrinet
= 32x FibreChannel
MyrinetClos
Spine Spine MyrinetClos
Chicago & LA DTF Core Switch/Routers Cisco 65xx Catalyst Switch (256 Gb/s Crossbar)
= 8x FibreChannel OC-12 OC-12 OC-3 vBNS Abilene MREN Juniper M40
1176p IBM SP Blue Horizon OC-48 NTON 32 24 8 32 24 8 4 4 Sun E10K 4 1500p Origin UniTree 1024p IA-32 320p IA-64 2 14 8 Juniper M40 vBNS Abilene Calren ESnet OC-12 OC-12 OC-12 OC-3 8 Sun Starcat 16 GbE
= 32x Myrinet
HPS S 256p HP X-Class 128p HP V2500 92p IA-32 24 Extreme Black Diamond OC-12 ATM Calren 2 2
Small Devices Increasing in Importance
n There is growing
interest in wireless portable displays in the confluence of cell phone and personal digital assistant
markets
n By 2005, 60 million
internet ready cell phones sold each year
n 65% of all
Broadband Internet accesses via non
desktop appliances
CM5
Integration of PDA’s and supercomputers (etc.) implies very heterogeneous
systems spanning
traditional performance fields
The HPCC Thrust has run its course?
n
The
1990 HPCC 10 year initiative
was largely aimed at
parallel computing enabling large scale simulations for a
broad range of computational science and engineering
problems
n
It was in many ways a success and we have methods and
machines that can (begin to)
tackle most 3D simulations
•ASCI simulations particularly impressive
•
DoE still putting substantial resources into basic software
and algorithms from adaptive meshes to PDE solver
libraries
n
Machines are still increasing in performance exponentially
and should achieve
petaflops
in next 7-10 years
n
Not obvious that there will be major changes in parallel
e-Science
n
e-Science
implies integration of
data and researchers around the
model and builds on
• Parallel Computers for Simulation • Sensors (satellites or ground based)
for data • Database
for knowledge • Networks to lin
Classic Grid Architecture
Database Database
Netsolv e
Computin g
Securit y Collaboratio
n
Compositio n
Content Access
Resources
Middle Tie Brokers Service Providers
Astronomy is Facing a
Major Data Avalanche
Astronomy is Facing
a Major Data
Avalanche:
Multi-Terabyte Sky Surveys and Archives (Soon: Multi-Petabyte), Billions of
Detected Sources, Hundreds of Measured Attributes
per Source …
Total area of 3m+ telescopes in the world in m2, total number of CCD pixels in Megapix, as a function of time. Growth over 25 years is a factor of 30 in glass, 3000 in pixels.
One e-Science Example
The Changing Style of Observational
Astronomy
Virtual
Observatory
Archives of pointed observations
(~ TB) Small samples
of objects
(~ 101 - 103)
Multiple, federated sky surveys and archives (~
PB) Large, homogeneous sky
surveys
(multi-TB, ~ 106- 109
sources) Pointed,
heterogeneou observations (~ MB - GB)
Future:
Now:
What is the NVO? - Content
Source Catalogs Image Data
Query Tools
Specialized Data:
Spectroscopy, Time Series,
Polarization Information Archives:
Derived & legacy data: NED,Simbad,ADS, etc
Analysis/Discovery Tools:
Visualization, Statistics
Standards
What is the NVO? - Components
n
Information Providers
e.g. ADS, NED, ...
Data Providers
Surveys, observatories, archives, SW repositories
Service Providers
Grid/P2P Use of Internet I
ROBERT B. COHEN, PH.D. COHEN COMMUNICATIONS GROUP [email protected] 212-986-7720
Global Grid Forum Toronto Feb 18 2002
Cohen’s Rival Estimate Mainl
Digital Video
Grid/P2P Use of Internet II
S2S Server to Server
Digital Vide “on demand”
P2P Grid
Use of Object Technologies I
n
The claimed commercial success in using
Object and
component technology
has not
yet
been a clear success in
HPCC and indeed in modeling & simulation
•
Object technologies
do not naturally support either
high performance or parallelism
•
C++
can be high performance but
Java (as a language)
is not uniformly so (it is improving)
•
We suggest that
Web Services
could change this
n
Fortran
(including Fortran90) will continue to decline in
importance and interest – the community should prefer
not to use it
•
It’s use will not attract the best students
n
Not essential
to write modules in
object oriented language
•
It is
essential
to package modules in
object framework
Use of Object Technologies II
n
There is
emerging HPCC component architecture
allowing
production of more modern libraries (integration
Infrastructure)
•
DoE has very large
CCA
– Common Component
Architecture – effort
•
Package software (“system and applications”)
as
distributed objects
– not as traditional libraries
n
CORBA HLA Java
and
Web Services
are
not
naturally
high
performance as
component models
•
High performance
often
not essential
for
coarse grain
objects
•
Web Services
support multiple implementations
allowing
Object Size & Distributed/Parallel Simulations
n
All
interesting systems
consist of
linked entities
•
Particles, grid points, people or groups thereof
n
Linkage translates into
message passing
•
Cars on a freeway
•Phone calls
•
Forces between particles
n
Amount of communication
tends to be proportional to
surface area of entity whereas simulation time proportional
to volume
n
So
communication/computation
is surface/volume and
decreases
in importance as
entity size increases
n
In parallel computing, communication synchronized; in
distributed computing “self contained objects” (whole
programs) which can be scheduled asynchronously
Some Problem Classes
n
Classic HPCC:
synchronized objects with regular time
structure (communication overhead decreases as
problem size increases)
• Includes PDE and interacting particle based applications
• Give scaling parallelism on large MPP’s
n
Grid: Internet Technology and Commercial Application
Integration:
Large objects with modest communications
and without difficult time synchronization
• Compose as independent (pipelined) services
• Includes some approaches to multi-disciplinary simulation
linkage
n
Hardest:
smallish objects with irregular time
synchronization
• Event driven simulations (HLA-RTI) used here
Sets of Grid Points
Sets of Service (programs)
What is a Web Service I
n A web service is a computer program running on either the local
or remote machine with a set of well defined interfaces (ports) specified in XML (WSDL)
n In principle, computer program can be in any language
(Fortran .. Java .. Perl .. Python) and the interfaces can be implemented in any way what so ever
• Interfaces can be method calls, Java RMI Messages, CGI Web
invocations, totally compiled away (inlining) but
n The simplest implementations involve XML messages (SOAP)
and programs written in net friendly languages like Java and Python
n Web Services separate the meaning of a port (message) interface
from its implementation
n Enhances/Enables Re-usable component model of ANY
electronic resource
What is a Web Service II
n
Web Services have important implication that
ALL
interfaces are XML messages based.
In contrast
n
Most Windows programs have interfaces defined as
interrupts due to user inputs
n
Most software have interfaces defined as methods which
might be implemented as a message but this is often
NOT explicit
Securit
y Catalog
Paymen Credit
Card
Warehous
etc. XML WS to WS Interfaces
(Virtual) XML Knowledge (User) Interface
Clients
(Virtual) XML Data Interface Raw Data
Ra
Resource
s
Raw Data W S W S Web Service (WS) W S W S WS WS WS
W S
Details of WSDL Protocol Stack
n
UDDI
finds where programs are
•
remote( (distributed) programs
are just Web Services
n
WSFL
links programs togethe
(under revision?)
n
WSDL
defines interface (methods,
parameters, data formats)
n
SOAP
defines structure of message
including serialization of information
n
HTTP
is negotiation/transport
protocol
n
TCP/IP
is layers 3-4 of OSI
Physical Network
is layer 1 of OSI
UDDI or WSIL
WSFL
WSDL
SOAP or RMI
HTTP or SMTP or IIOP or RMTP
TCP/IP
Message Or Event Based Inte
Connection
Reso urce
Data base
Reso urce Sof
ware
Sof ware
XM Skin
e-Science/Grid/P2P Networks are XML
Specified Resources connected by XML
specified message
Implementation of resource and connection may or may not be XML XM
Skin
What is a Grid Web Service?
n There are generic Grid system services: security, collaboration,
persistent storage, universal access
• OGSA (Open Grid Service Architecture) is implementing these as
extended Web Services
n An Application Web Service is a capability used either by another
service or by a user
• It has input and output ports – data is from sensors or other
services
n Consider Satellite-based Sensor Operations as a Web Service • Satellite management (with a web front end)
• Each tracking station is a service
• Image Processing is a pipeline of filters – which can be grouped
into different services
• Data storage is an important system service
• Big services built hierarchically from “basic” services
Sensor Web Service
Distributed Sensor Web Service
Output Web Service port
Universal sensor acces for people/computers
Input Web Service port
Different forma Sensor Data
Application Web Services
n Note Service model integrates sensors, sensor analysis, simulations and people n An Application Web Service is a capability used either by another service or
by a user
• It has input and output ports – data is from users, sensors or other services • Big services built hierarchically from “basic” services
Sensor Data as a We
service (WS) Data Analysis WS Sensor Managemen WS Visualization WS Simulation WS Filter
WS FilterWS FilterWS
Build as multiple Filter Web Services
Prog
WS ProgWS
The Application Service Model
n
As bandwidth of communication (between) services increases
one can support smaller services
n
A service “is a
component
” and is a replacement for a
library in case where performance allows
n
Services (components)
are a sustainable model of software
development – each service has documented capability with
standards compliant interfaces
•
XML
defines interfaces at several levels
•
WSDL
at Service interface level and
XSIL
or equivalent
for scientific data format
n
A service can be written as Perl, Python, Java Servlet,
Enterprise Javabean, CORBA (C++ or Fortran) Object …
n
Communication
protocol can be RMI (Java), IIOP
(CORBA) or SOAP (HTTP, XML) ……
Some Science Web Services
n
These build on general (community) web services
Education as a Web Service
n Can link to Science as a Web Service and substitute educational
modules
n “Learning Object” XML standards already exist from IMS/ADL
http://www.adlnet.org – need to update architecture
n Web Services for virtual university include:
n Registration
n Performance (grading) n Authoring of Curriculum
n Online laboratories for real and virtual instruments n Homework submission
n Quizzes of various types (multiple choice, random parameters) n Assessment data access and analysis
n Synchronous Delivery of Curricula
n Scheduling of courses and mentoring sessions
n Asynchronous access, data-mining and knowledge discovery
Different Web Service Organizations
n
Everything is a
resource implemented as a Web
Service
, whether it be:
•
back end supercomputers and a petabyte data
•
Microsoft PowerPoint and this file
n
All
Resources communicate via messages
n
Grids
and
Peer to Peer (P2P) networks
can be
integrated by building both in terms of
Web
Services
with different (or in fact sometimes the
same) implementations of
core services
such as
registration
,
discovery
,
life-cycle
,
collaboration
and
event or message transport
…..
•
Gives a
Peer-to-Peer Grid
Peer to Peer Grid
Database Database
JXTA
JXTA
Web Service Interfaces
Web Service Interfaces
Event Messag Brokers
Integrate P2P and Grid/WS
Role of Event/Message Brokers
n We will use events and messages interchangeably
• An event is a time stamped message
n Our systems are built from clients, servers and “event brokers”
• These are logical functions – a given computer can have one
or more of these functions
• In P2P networks, computers typically multifunction; in Grids
one tends to have separate function computers
• Event Brokers “just” provide message/event services; servers
provide traditional distributed object services as Web
services
n There are functionalities that only depend on event itself and
perhaps the data format; they do not depend on details of application and can be shared among several applications
• NaradaBrokering is designed to provide these functionalities
• MPI provided such functionalities for all parallel computing
NaradaBrokering implements an
Event Web Service
n Filter is mapping to PDA or slow communication channel
(universal access) – see our PDA adaptor
n Workflow implements message process n Routing illustrated by JXTA
Destination-Source matching illustrated by JMS using
Publish-Web
Service 1 (VirtualQueue Service 2Web
Destinatio
Source Matching Filter
Routin
g workflow
WSD
Features of Event Service I
n
MPI
nowadays aims at a
microsecond latency
n
The Event Web Service aims at a
millisecond latency
• Typical distributed system travel times are many milliseconds
(to seconds for Geosynchronous satellites)
• Different performance/functionality trade-off
n
Messages are
not sent directly
from
P
to
S
but rather
from
P
to Broker
B
and from Broker
B
to subscriber
S
n
Synchronous
systems: B acts as a real-time
router/filterer
• Messages can be archived and software multicast
n
Asynchronous
systems: B acts as an
XML database
and
workflow
engine
n
Subscription is in each case, roughly equivalent to a
database query
Features of Event Web Service II
n
In principle Message brokering can be virtual and
compiled away
in the same way that WSDL ports can
be bound in real time to optimal transport mechanism
• All Web Services are specified in XML but can be
implemented quite differently
• Audio Video Conferencing sessions could be negotiated using SOAP (raw XML) messages and agree to use certain video codecs transmitted by UDP/RTP
n
There is a collection of XML Schema – call it GXOS –
specifying
event service
and requirements of message
streams and their endpoints
• One can sometimes compile message streams specified in
GXOS to MPI or to local method call
n
Event Service must support dynamic heterogeneous
Features of Event Web Service III
n
The event web service is naturally implemented as a
dynamic distributed network
• Required for fault tolerance and performance
n
A new
classroom joins
my online lecture
• A broker is created to handle students – multicast locally my
messages to classroom; handle with high performance local messages between students
n
Company X sets up a
firewall
• The event service sets up brokers either side of firewall to
optimize transport through the firewall
n
Note
all message based applications
use
same message
service
• Web services imply ALL applications are (possibly virtual)
message based
Broker Network
Data base Reso
urce
Broker
Broker Broker Broker
Broker
Broker
Software multicast
(P2P) Community
For message/events service (P2P) Community
System Structure I
n
Systems are a dynamic mix of structured and
unstructured entities
n
P2P systems like
JXTA
support unstructured systems
realized by opportunistic messaging “broadcast locally”
over a certain “network distance”
n
Java Message Service
JMS
supports structured systems
where clients (message endpoints) link to one of a
known set of “central servers”
n
Event system must support
• Advertise capability – Publish
• Advertise need – Subscribe both for type and form of messages
• Transport designated messages/events
Single Server P2P Illusion
Collaboration Server Data
base
System Structure II
n
One could think that the world is a well defined
structure of unstructured systems
• Unstructured dynamic systems are P2P (JXTA) Peer Groups
• Peer Groups could be cluster of students in a class for
distance learning or cluster of Grid (OGSA) Web services generated to support running a job
n
But maybe it is a set of
structured communities
with
unstructured connection
n
NaradaBrokering
needs to support both models and
those in between
• Currently has JMS mode, JXTA mode and Native (most
powerful) mode
n
P2P
usually thought of as a set of “
unruly dangerous
clients
” but can equally well be used securely as a
middleware
interaction mode between
web services
Database Database
Grid Middleware
Grid Middleware
Grid Middlewar
e
Grid Middlewar
e
MP Group MP Group
M
p
M
Community Grids Laboratory Activities I
n
Core NaradaBrokering
Event Service
• Operation in JMS or JXTA mode to demonstrate integration
of central and peer-to-peer mode
• Focus is Performance and Capabilities (see later)
n
Garnet synchronous collaboration
environment used
for distance education and seminars
• Built first on commercial JMS but ported to Narada – shows
that one can afford to use message service in synchronous application sharing
n
Interface of
Garnet
to
PDA
with
message size filtering
and optimized
HHMS
message service
• This filtering also needed for slow clients – mix of dial-ups
and Internet2 clients in a collaboration
• Event system supports (XML) client profiles
NaradaBrokering and JMS
Low Rate; Small Messages
NaradaBrokering and JXTA
Comparing Pure JXTA, Narada-JXTA and Direct P2P There is a bug in JXTA and this was only just fixed
Narada-JXTA provides JXTA guaranteed long distance delivery
Small Payload
JXTA is getting slower
Pure Narada 2 hops
Client
Client Narada
Narada
Client JXTA JXTA Client
Client JXTA Narada JXTA Client
Client JXTA JXTA Client multicast
PowerPoint can be converted to SV via Illustrator or Web export
Batik Viewer on PC
PDA Collaboration Event Filter
GMS JMS o
Narada This nowDoing
GMSME : iPaq H3650, WinCE 3.0,
Personal-Java1.1 Wireless 11 Mbit/s IEEE 802.11b
Community Grids Laboratory Activities II
n
Use of
JMS
(Narada) to support
asynchronous
collaboration
including early GXOS Schema XML
based
News Groups
and Web Site management
• Integrated with Apache Slide and Jetspeed portals
n
Audio-Video Conferencing
as a Web service
• H323 and SIP as Web services using XML Session Schema
• NaradaBrokering support of UDP
n
Computing Portals
as Web services; NaradaBrokering
could support events (status, performance, job flow)
linking
operational job
to control servers and
NaradaBrokering Futures
n
Higher Performance
– reduce minimum transit time to
around one millisecond
n
Substantial
operational testing
n
Security
– allow Grid (Kerberos/PKI) security
mechanisms
n
Support of more
protocols
with dynamic switching as
in JXTA – SOAP, RMI, RTP/UDP
n
Integration of simple
XML database model
using JXTA
Search to manage distributed archives
n
More formal specification of “
native mode
” and
dynamic instantiation of brokers
n
General Collaborative Web services
Collaborative Web Service Access
n Intercept and multicast messages produced by Web Service
Collaborative We Service
Maste Client
Clien t
Even (Message
Service
Web Service has a por on which collaborativ modes set
Web Service can b “front-end” (in middl tier) to complex
Web Service Intercepto Providing General Services
Set Collaboration and Message Mode
Collaboratio as a We
Collaborative Replicated Web Services
n Intercept and multicast messages SENT to Web Service
Objec t Object Displa y Object Viewer Object Displa y Object Viewer Even (Message Service Master Object Displa y Object Viewer We Servic e Web Service Intercepto
Providing General Services Set Collaboration Mode