Rulebased SLA mediation

1
Rule-based SLA mediation

• Andras Micsik, Henar Muñoz Frutos

2
Problem Area

• RAMMemory 7.5 GB
• ComputeUnit 4 ECU
• Storage 850GB
• Platform 64 bit
• Price 0.4 per instance hour

Different languages, different metrics
Negotiation
AMAZON

• ANSYS request
• CPUName IntelCore Duo
• CPU Speed 2 GHz
• Capacity 400 MB
• Price 27 euros per day
• DiskSpace 250GB

Negotiation

• MemoryPerTask 7.5 GB
• ClockCPUSpeed 100 MHz / process
• StorageCapability 850GB
• Cost 5 euros/task/hour

CUSTOMER
Software engineering company
BSC
2
3
Architecture
Comon conceptual model
SA-SLA(WS-AgreementWSLA)
SLA Mediator
SLA Mediator
SLA framework
SLA framework
Local domain knowledge
Local domain knowledge
4
Use of local mappings
Common SLA QoS ontology
Units
Metrics
MinimumCacheSize
MB
local mapping2
local mapping1
CacheSizeMin
MinCacheSize
negotiation
kB
MB
GuaranteeTerm
GuaranteeTerm
4096
4
Service Consumer
Service Provider
5
Main concepts of common ontology
Quality Factoror SLO
Measurement Unit
Quality Modelor SLA
Metric
BusinessQoS
Offer
TimeUnit
InfrastructureMetric
Request
ImplementationQoS
StorageMetric
StorageUnit
Agreement
InfrastructureQoS
MemoryMetric
MonetaryUnit
PerformanceMetric
ApplicationQoS
ResponseTime
NetworkQoS
PriceMetric
6
Organization of BREIN ontologies
S-BPMN
QoS Ontology
Business processes
SLAs
Business Ontology
Business layer
Implementation layer
OWL-S
OWL-WS
IT workflows
Services
Technical ontology
GRO
S-OGSA
Semantic bindings
Grid resources
7
Implementation of SLA Mediator

• Local knowledge is collected in OWL
• SLA templates,
• Available resources, resource types
• Mapping between common model and local model is
  provided in the form of
• Conversion definitions (e.g. rates in OWL)
• Conversion rules (in SWRL)
• Translation of incoming SLA bids is done as a
  sequence of
• Conversion of XML data into OWL data
• Conversion of common OWL data to local OWL data
  (i.e. using local metrics)
• Matching and ranking SLA bid with SLA templates
• Implementation uses a mixture of custom Java
  code, SWRL and OWL DL reasoning

8
Local interpretation of SA-SLA
RAMMemory
type
ltwslaSLAParameter nameCapacity"
type"double" unit"MB" sawsdlmodelReference"R
AMMemory"gt ltwslaMetricgtmemorylt/wsla
Metricgt lt/wslaSLAParametergt

slaParameter1
name
unit
GB
Capacity
RAMMemory
SLO1
slaParameter1
SLO1
localunit
type
value
gt
slaParameter1
gt
unit
kB
4
localvalue
name
value
unit
GB
GB
Capacity
4
4096
9
Evaluation of offers vs. bid
SLO3
Offer1
SLO2
1.0
SLO4
Bid1
0.5
SLO5
SLO1
Offer2
SLO6
10
Evaluation of SLA Mediator

• Advantages
• Incoming data is converted and processed in a
  localized format, applying locally used metrics
  and measuring units
• The matching and ranking of suitable templates is
  customizable to match specific local needs
• Using rules helps to avoid changing code
• Challenges
• Reasoning technology is slow
• Expressivity of SWRL is weak
• Ongoing work
• Matching SLA requests with current availability
  of resources
• Improving performance of reasoning

11
Issues for discussion

• Already several terminologies exist WSQM, WSLA,
  TMF, ... etc.
• How to harmonize these in a common QoS ontology?
• Which terminology to use to name concepts in a
  common QoS ontology?
• Performance and maturity problems of OWL
• Do we need safety or non-monotonicity with rules?
• Getting good GUIs for OWL and rules management
• Do we need slow OWL or fast RDF repositories?