Ontologybased Resource Matchmaker

1
Ontology-based Resource Matchmaker

• Hongsuda Tangmunarunkit
• Carl Kesselman
• Center for Grid Technologies

Stefan Decker Intelligent System Division
Information Sciences Institute University of
Southern California
2
Current Execution Environment
3
Pathway 2 Workflow

• Physics-based EQ simulation Workflow
• Transfer source definition to almaak.usc.edu
• Run CVM code on almaak.usc.edu
• Transfer CVM output to hpc.usc.edu
• Run Olsen simulation on hpc.usc.edu
• Transfer outputs to pinto.isi.edu
• Run visualization tool on pinto.isi.edu

Visualization tools
Anelastic Wave Propagation Simulation
Problem definition
Velocity Model Code and Data
6
Geometric and physical properties and max freq
1
3
5
2
4
Olsens code
CVM (CMU)
Outputs
Input mesh
Outputs
Storage System
Source definition (e.g. fault locations)
pinto.isi.edu
almaak.usc.edu
hpc.usc.edu
pinto.isi.edu

• With direct scripting, we have successfully run a
  Northridge earthquake simulation for San
  Fernando basin.

4
Ontology-based Resource Matchmaker
E.g., Request1 (owneruser1) (JobTypeMPI)
(NumberCPUs48) (Memory1000) (Disk10000)
E.g., Matches Request1 with LinuxCluster.usc.edu
(AvailableCPUs100) (Memory1500) (Disk30000)
5
Ontology-based Matchmaker (OMM)

• Semantic Matching
• Separate ontologies (i.e., semantic descriptions)
  of resources and requests
• Semantic matching by using rules based on terms
  defined in ontologies
• Flexible extensible new concepts can be added
  by adding new vocabulary and inference rules
• Based on Semantic Web Technologies
• Standard Ontology Language RDF-Schema
• Rule Language Triple
• Support RDF-schema
• Deductive Database System Triple/XSB

6
Desired Features

• Asymmetric description of resource and request
• Expressiveness
• Allow users to describe request in terms of
  high-level application characteristics, e.g., MPI
  application.
• Bilateral Constraints
• Resource expresses its usage policies restricting
  matches to applications/requests
• Request specifies constraints in terms of
  requirements
• Ability to specify matching preference
• Multi-lateral matching a collection of resources
• Gang matching A request describing each resource
  individually
• Set matching a request in terms of aggregate
  properties (e.g., a total disk space of 100GB)

7
Matchmaker Architecture
Advertisements/ Requests
Errors/Reply

• - Matching Criteria (e.g.,
• Request.OS must be
• compatible w/ Resource.OS)
• - Matchmaking Algorithm
• (e.g., return the highest rank
• compatible resource)
• Integrity Checking (e.g.,
• check for consistency)

Ontology-based Matchmaker
Background knowledge about the domain, e.g.,
SunOS, Linux are types of Unix
operatingSystem
Domain Models objects, their properties
relationships between objects (e.g.,
ComputerSystem, OperatingSystem, RunningOS)
Domain Ontologies (Resources,Policies,Requests)
Deductive Database System (Triple/XSB)
8
Ontologies

• Developing 3 Ontologies using RDF-Schema
• Resource Ontology
• Describe resource components, their properties
  and relationships
• Classes ComputerSystem, OperatingSystem
• Relationships RunningOS
• Based on Common Information Model (CIM)
• Resource Request Ontology
• Describe applications, characteristics and
  resource requirements, e.g., JobTypeMPI
• Policy Ontology
• Resources specify their usage and authorization
  policies

9
Resource Ontology
Currently focus on Computation Resources
10
Domain Background Knowledge

• Capture background knowledge about the domain
  (usually at the instance level) in terms of
  rules.
• SunOS and Linux are types of Unix.
• Can substitute Unix with SunOS.

_at_gridBackground // grid background knowledge
SunOStype?Unix. Linuxtype?Unix. //
Compat Transitivity axiom FORALL X,Y,Z
XCompat?Z? Xcompat?Y AND Ycompat?Z.
// Substitution rule FORALL X,Y,Z
Xsub?Z ? (Ytype?Z and Xsub?Y) or
Xcompat?Z.
11
Matchmaking Rules Matching
FORALL Data, Background _at_match(Data,Background,Ont
ology) FORALL X,Y Xmatches?Y ?
Xtype?Request and Ytype?Resource and
X.RequiredFSmatchesFS?Y.HostedFileSystem
and X.RequiredOSmatchesOS?Y.RunningOS . //
check FS Constraint FORALL X,Y XmatchesFS?Y
? Xtype?FSRequirement and Ytype?OperatingSyste
m and X.MinDiskSpace lt Y.AvailableSpace
. // check OS constraint FORALL X,Y
XmatchesOS?Y ? Xtype?OSRequirement and
Ytype?OperatingSystem and
X.OSTypesub?Y.OSType .
Simple arithmetic/string comparison
Reasoning based on background knowledge and class
hierarchy
12
Example
Resource
Request
FORALL X,Y lt- Xmatches-gtY_at_match(advertisement,gr
idBackground,gridOntology). X Request1, Y
Almaak.usc.edu
Query
13
Future Work

• Ontology-based matchmaker
• Performance Evaluation (in terms of resources
  and rules)
• Ontology-based matchmaker as a Grid service
• Expand ontology vocabularies (e.g., to support
  other resources and services)
• Develop high-level application description
• Investigate Olsens Wave Propagation Simulations
• Example Request1 (Olsen01 EarthVolume
  Timestep)
• Integration matchmaker with other components
  (i.e., Docker, Pathway Ontology, Digital
  Libraries, Pegasus, Chimera)

14
Future Execution Environment
Grid Client
Users request
Knowledge Acquisition (Docker)
SCEC Pathway Ontology
Resource Data Information
Workflow Manager
Job Submission Manager (Condor-G)
Grid Resource Ontologies
OMM
Knowledge Sources
Grid Information Services