Post on 25-Jan-2016
description
SZTAKI Desktop Grid
P. Kacsuk
MTA SZTAKI
www.lpds.sztaki.hu
Desktop Grid model
Internet/Intranet
Dynamic resource donation
Work package distribution
Company/univ.
server
Donor: Company/
univ. or private PC
Donor: Company/
Univ. or private PC
Donor: Company/
univ. or private PC
Application
Characteristics of the desktop Grid model
• Anybody can donate resources• Heterogeneous resources, that dynamically join
and leave• One or a small number of projects can use
the resources• Asymmetric relationship between donors and
users: U << D
• Advantage: • Donating a PC is extremely easy• Setting up and maintaining a DG server is much easier
than installing the server sw of utility grids• Disadvantage:
• Dynamic job submission is not supported• Supported application is static (typically very long
running application)
Master/slave parallelismand parameter studies
Internet
Master
Workunit-1
Workunit-2
Workunit-3
Workunit-N
DG Server
Types of Desktop Grids
• Global Desktop Grid• Aim is to collect resources world-wide for grand-
challenge scientific problems
• Examples: • SETI@home• SZTAKI Desktop Grid global version at:
http://szdg.lpds.sztaki.hu/szdg/
• Local Desktop Grid• Aim is to enable the quick and easy creation of
grid for any community (company, univ. city, etc.) to solve their own applications
• Example: • SZTAKI Desktop Grid local version
Firew
all
Firew
all
Global Desktop GridsGlobal Desktop Grids
NATNAT
clients join the server voluntarily public resources and a public project resources can be attached to more Desktop Grid projects resources are mostly behind Firewalls or NAT resources come and go, fluctuating performance
Local Desktop GridsLocal Desktop Grids
Institute/ Dept.
private (dedicated) resources, private project(s) more freedom for applications choice and administration oriented for companies and institutes more predictable performance
Firew
all
SZTAKI Desktop Grid: Global version
• Main objective: • Demonstrate the usage of DG systems for
any community
• Supported application: searching 12-dimension binary number systems
• Number of registered donors: >19000• Number of registered PCs: > 35000• How to register a PC?
http://www.lpds.sztaki.hu/szdg/
SZTAKI Desktop Grid global version
SZTAKI Desktop Grid performance
TOP 500 entry performance: 1645 GFlops
SZTAKI Desktop Grid donors (users)
SZTAKI Desktop Grid local version
• Main objective: • Enable the creation of a local DG for any
community• Demonstrate how to create such a system
• Building production service Grids requires huge effort and represents a privilege for those organizations where high Grid expertise is available
• Using the local SZDG package • Any organization can build a local DG in a day with
minimal effort and with minimal cost• The applications of the local community will be executed
by the spare PC cycles of the local community• There is no limitation for the applied PCs, all the PCs of
the organization can be exploited (heterogeneous Grid)
Usage of Sztaki Desktop GridUsage of Sztaki Desktop Grid
DSP application on a local SZDG in the Univ. of
Westminster• Digital Signal Processing
Appl.: Designing optimal periodic nonuniform sampling sequences
• Currently more than 100 PCs connected from Westminster and planned to extend over 1000 PCs
DSP size Production SZDG
20
22
24
~35min ~1h 44min
~7h 23min
~141h ~46h 46min
The speedup
~5h 4min
Sequential
~3h 33min
~41h 53min
~724h
Usage of local SZDG in industry
• Comgenex Ltd. -> AMRI Hungary• Drug discovery application• Creating enterprise Grid for prediction of
ADME/Tox parameters• Millions of molecules to test according to potential
drug criteria
• Hungarian Telecom• Creating enterprise Grid for supporting large data
mining applications where single computer performance is not enough
• OMSZ (Hungarian Meteorology Service)• Creating enterprise Grid for climate modeling
● Normal Desktop Grid (the current SZDG, available as package)
● Desktop Grid with cluster extension (available as package)
● Goal is to include clusters in an SZDG (EU CancerGrid Project)● Hierarchical Desktop Grid (available as prototype)
● Goal is to build larger DGs from smaller ones in a hierarchical way
● E.g. Enterprise DG can be built exploiting PCs of the dept. DGs● Expected release by October 2007 (Hungarian DG project)
Components of SZDG
LocalDEG
LocalDEG LocalDEG
Normal Desktop Grid
University Dept. DG
University Faculty DG
• Each local DG runs the applications of the local community (univ. dept., faculty, enterprise, etc.)
Enterprise DG
LocalDEG
LocalDEG LocalDEG
Mixed Desktop Grid
University Dept. DG
University Faculty DG
Enterprise DG
Local DGs can be extended with local clusters
LocalDEG
LocalDEG
LocalDEG
LocalDEG LocalDEG
Hierarchical Desktop Grid
University DG
Enterprise Dept. DG
• Local DGs at the lower level of hierarchy can be used to solve the applications of the higher level DGs.
• E.g. univ. dept. and faculty DGs contribute to the university level DG
University Dept. DG University Faculty DG
Enterprise DG
HierarchyHierarchy
DG projects need modifications only for app registration
ChallengesChallenges
based on the prototype we implemented
redundant workunitsredundant workunits
application representationapplication representation
application registrationapplication registration
trusttrust
workunit deadlinesworkunit deadlines
Redundant workunitsRedundant workunits
redundancy can be enabled only on the highest level
it is automatically disabled on every other level
Workunit deadlinesWorkunit deadlines
deadlines ensure that no workunit can be hijacked deadline is set when downloaded
problem in hierarchy workunits transferred to lower levels have the
deadline “timer” ticking
workunit download management required
TrustTrust
public/private keypair for code signing and workunit signing
that’s not enough for the hierarchy
From where should I accept applications and work ?From where should I accept applications and work ?
Who/ what guarantees that the app does no harm ?Who/ what guarantees that the app does no harm ?
How can I trust the work provider ?How can I trust the work provider ?
TrustTrust
to solve the problem we introduce X.509 certificates
not just for code signing and wu signing, but also for distinguishing
the Application Developerthe Application Developer
the Projectthe Project
the Serverthe Server
the Clientthe Client
Application representationApplication representation
the application has a name and a version the Application Developer signs the
application
the application is identified by the name, version, signature
Application registrationApplication registration
Application registrationApplication registration
If we trust the Application Developer, we also trust the application signed by her
the project has a CA List with the list of trusted Application Developers and Projects
if the Application Developer is trusted, the application is deployed
The Project may sign the application
The application registration processThe application registration process
signsignsignsign
installinstall
attachattach
contact downloadcert, CA list
downloadcert, CA list
wu querywu query
get appget appwu querywu query
deploydeploy
Conclusions of hierarchical SZDGConclusions of hierarchical SZDG
Hierarchy is possible with minimal modification of existing projects
For increased security or industrial usage certificates provide a good solution
By introducing certificates most of the limitations can be solved
The hierarchical version of SZDG works at prototype level and will be released during this year
● DC API: available in the current SZDG as package (Hungarian DG project)
● Goal is to provide an easy-to-use library for generating Master/worker type Grid applications both for DG and SG systems
● See the next lecture by Gabor Gombas● Portal access to SZDG: (in planning phase
within the EU CancerGrid project)● Goal is to provide a high-level, graphical workflow
level portal to generate and run SZDG applications
Application development support in SZDG
Portal support for SZDG
● Disadvantage of BOINC:● Creation BOINC applications is a difficult task● A BOINC application is static, not submittable
● Our goal:● Enable the dynamic submission of workflow
and parameter sweep applications into the SZDG system
● Make it easy to create and submit workflow and parameter sweep applications for SZDG
An example CancerGrid workflow
x1
x1
x1
xN
xN
NxM
NxM
NxM
xN
xN
xN
xN
NxM
Generator job Generator job
N = 20e-30e, M = 100N = 20e-30e, M = 100 => very large number of executions and files => very large number of executions and files
P-GRADE Portal Submitter Service
● A web service component of the portal● Receives job submission requests from the
portal workflow manager● Must be prepared to do the following
tasks:● Submit the job● Check the submitted job’s status● Get the job’s output● Abort the job
Submitter - Overview
P-GRADE Portal
A
B C
D
WF Manager
GRID
SubmitterService
BSubmit
B Submit,Check stat,
Abort,Get output
Reportstatus
change
Portal – SZDG integration tasks
● Create a special submitter for BOINC● Manipulate BOINC database: workunit
creation, result status query, …● Glue together short-running algorithm
instances● Consider job and job owner priorities if needed
● Goal: minimize overhead of BOINC, minimize network traffic, but do not increment response time to users
CancerGrid DBMS SQL
Portal + SZDG (BOINC)
Portal
BOINC Szerver
PortalStorage
Submitter
Queue Manager + Scheduler
BOINC DB - MySQL
Q1
Qn
Donor
Donor
Donor
WU WU WU
Q2
2D3D
Mopac
• The submitter stores the jobs sent by the portal in algorithm queues• Once a queue contains „enough” jobs, a BOINC workunit is created
Solution – Short-running jobs
● Instances of algorithms running for a short time should be glued together in order to
● increase computation time ● and minimize BOINC overhead
● Sheduling questions:● How many jobs are „enough”?● How long should we wait for new jobs?● What about algorithm/user priorities?● Detect „bad” donors?
● A Queue Manager takes care of algorithm queues
Enabling Desktop Grids for e-Science(EDGeS)
● New FP7 project to be started on the 01/01/2008● Goals of the project:
To integrate Service Grids (SG) and Desktop Grids (DG) to attract new scientific communities that needs very large number of computing resources
To involve new type of user and resource provider communities beyond the scientific communities (school students, citizens of cities, companies)
To provide APIs and Grid application development tools for the new scientific user communities in order to adapt their applications for the integrated SG-DG e-infrastructure
To adapt the identified applications for the integrated SG-DG e-infrastructure
To provide new trust mechanisms for the integrated SG-DG e-infrastructure
To establish international collaborations, procedures and standardsTo contribute to the establishment of a sustainable Grid
infrastructure in Europe● The infrastructure to be established:
● Integrated Service Grid – Desktop Grid system
Sevice Grid (EGEE)
Public DG
SZDG
30.000 PCs
Local DG UoW Grid
1.500 PCs
Public DG IN2P3 Grid
300 PCs
Public DG EGEE-BOINC
Planned 10.000 PCs
Public DG Extremadura
Grid
70.000 PCs
Public DG EGEE-
XtremWeb
1.000 PCs
Public DG AlemereGrid
3.000 PCs
BOINC based DGs
Local DG IN2P3 Grid
200 PCs
XtremWeb based DGs
The EDGeS infrastructure
LocalDEG
LocalDEG
LocalDEG
LocalDEG LocalDEG
Production Grid EGEE
Variety of DG systems within EDGeS
University DG
Enterprise DG
Public DG
SG (EGEE)
Local DG
Public DG
Public DG
Local DG
Local DG
Local DG
Public DG
Public DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Interoperability of EGEE with DG systems
SG (EGEE)
Local DG
Public DG
SG (SEE-Grid)
SG (EELA)
Public DG
Local DG
Local DG
Local DG
Public DG
Public DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Local DG
Towards unlimited Grid resources
Assessment of SZDG
• Advantages• Easy to create and maintain• Any organization can quickly and cheaply install it• Easy to program and hence no steep learning curve• Robust technology• Industry can use it as enterprise Grid• It extends BOINC with:
• Clusters as donated client systems• Hierarchy of DG systems• DC-API and portal access for easy generation of DG
applications
Conclusions
● Desktop Grids are here for any community (universities, companies , etc.). They can
● access and/or ● build Grid systems
● SZTAKI Desktop Grid technology enables ● easy creation and programming of global and local desktop
Grids● SZTAKI is ready to help any organization
● to set up its local DG(s)● to port applications on local DGs ● to train people how to build and use local DGs
● More information on:http://www.desktopgrid.hu/
http://www.lpds.sztaki.hu/szdg/