Riskaware Migrations for Prepossessing SLAs

1
Risk-aware Migrations for Prepossessing SLAs

• Kerstin Voß, Paderborn Universitykerstinv_at_upb.de

2
Motivation

• commercial users have a desired level of quality
  for job execution
• SLAs define relationship between end-user and
  provider(performance, charge and penalty fee)
• agreeing an SLA is a business risk for providers
• SLA is only a concept
• AssessGrid (EC-funded) integrate risk assessment
  and management into all Grid layers
• knowing the risk is a basis for providers
  decisions
• negotiations
• resource allocation
• fault-tolerance mechanisms

Migrations
Motivation
General Idea
Conclusion
3
Agenda

• Motivation
• Idea of Integrating Risk Assessment and
  Management
• Risk-aware Migrations
• Conclusion

4
Risk Assessment

• assess risk for SLA violation according to
• SLA definitions
• static and dynamic data of involved resources
• workload and available alternative resources
• available fault-tolerance mechanisms
• risk assessment new module in the RMS
• SLA defines
• time of data submission
• execution duration
• deadline
• resource requirements

Migrations
Motivation
General Idea
Conclusion
5
Using Risk Assessment

• compute risk for SLA violation before offering an
  SLA
• make a reservation
• publish the risk in SLA
• determine charge and penalty fee
• monitor risk of SLA violation during job
  execution
• initiate precautionary FT-actions if risk
  increases too much
• checkpointing and migration select a suitable
  resource for migration
• these actions are risk management!
• included in RMS modules

Migrations
Motivation
General Idea
Conclusion
6
Scheduler

• planning-based system
• make reservations when offering SLA
• for risk reduction insert an additional
  reservations for migrations (FT-reservation)
• different timeslots
• objectives balance schedule quality and risk for
  execution

reserved for j1
reserved for executing j2 if risk had increased
free
execute j3
schedule of resource i
Migrations
Motivation
General Idea
Conclusion
7
Decision of Selecting a Resource

• prime objective minimize risk for SLA
  violation? migration to resources with
  acceptable risk
• many free suitable resources ? similar to
  reservation
• high workload ? no/few suitable resources with
  adequate free timeslot
• compare effects of migration to resources with
  acceptable risk for execution
• damage can be measured in profit (penalty) or
  trustworthiness
• resulting lost of damage should be minimized

Migrations
Motivation
General Idea
Conclusion
8
Example Scenario
Migrations
Motivation
General Idea
Conclusion
9
Possibilities to Build a Free Timeslot

• time-shift job-reservations, FT-reservations, and
  planned jobs
• cancel FT-reservation
• cancel job-reservation
• cancel a planned job
• migrate a planned job to another resource
• combine several actions
• do not migrate

arbitrary
free
FT res.
exe. job
res. A
free
free
res. B
Migrations
Motivation
General Idea
Conclusion
10
Conclusion

• integrating risk assessment and management will
  be beneficial especially for the provider
• estimate the risk of agreeing an SLA
• precautionary migrations will prevent SLA
  violation
• selecting alternative has to consider
• risk of job execution
• resulting damage
• AssessGrid project www.assessgrid.eu

Thanks for your attention!
Migrations
Motivation
General Idea
Conclusion