Service Level Agreement Design and Service Provisioning for Outsourced Services

1
Service Level Agreement Design and Service
Provisioning for Outsourced Services

• Filipe T. Marques, Jacques P. Sauvé and Antão
  Moura
• Universidade Federal de Campina Grande

LANOMS 2007
2
The Problem
3
Context

• Service Level Agreement (SLA) Design
• How to choose SLA parameters?
• Service Provisioning
• How to design infrastructure to meet Service
  Level Objectives (SLOs)?
• This is an associated problem

4
Past solution BDIM

• We have a past solution using Business Driven
  IT Management (BDIM)
• Summary
• Invert!
• Dont go from SLOs to infrastructure design
• Design optimal infrastructure according to other
  criteria and then calculate SLOs
• Optimal design uses business considerations
• Tradeoff between cost and loss
• Minimize infrastructure cost losses due to
  performance
• Performance availability, response time
• Business Impact model links IT and the business

5
Problem with past solution

• Whose business are we talking about in past
  solution?
• Only the client's business
• Only applicable to in-house Service Providers
• ITIL v3 calls these Type I Service providers
• How to consider outsourced services?
• There are 2 businesses
• Sum of client loss and provider costs doesnt
  make sense
• SLA includes rewards and penalties, not
  considered in original model

6
Our Objective
7
Objectives

• Extend models and approaches to outsourced
  services
• Multidimensional design problem
• Choosing SLA parameters (availability, response
  time)
• Server farm design to provision the service
• Multi-objective optimization
• How to link Clients business and Providers
  business?
• Need new business-IT linkage models

8
Our solution
9
Conventional SLA design approach

• Choose SLOs (availability and response time) and
  penalties and calculate lowest-cost
  infrastructure to meet SLOs

10
Standard solution

• Build a performance model using available
  techniques
• Inputs
• Services, load, customer behavior
• Desired performance objectives (SLOs)
• Resource classes (tiers), resources and
  attributes (processing power, MTBF, MTTR, cost)
• Output
• In each class calculate how many load-balanced
  servers (for load) and standby servers (for
  availability)
• Cost of infrastructure and hence of service
• Our addition to standard models
• Performance model (reliability theory and queuing
  theory) to calculate response time distribution
  and availability

11
How to bring in business perspective?

• Better model the financial aspects of both
  businesses

12
Business model

• Service price markup over service price
• Provider revenue service price - penalties when
  availability and response time performance not
  met bonus as function of client profit
• Can use other provider revenue models
• Provider profit provider revenue - service cost
• Profit margin provider profit / provider
  revenue
• Loss due to unavailable service and poor
  response time
• Can calculate client revenue, cost, profit and
  profit margin
• Revenue model fixed revenue per transaction

13
Optimization problem

• Business-driven SLA design problem
• Maximizing service provider and client profit
  margins
• Multi-objective optimization problem
• This yields the optimal infrastructure
• Can simply calculate resulting SLOs from
  performance models

14
an example
15
The setting

• E-commerce
• 3 tiers web, application, database
• Customer Behavior Graph models page hits to
  browse catalog, ask for product details, buy,
  register, pay, etc.
• We use typical values for MTBF, MTTR, resource
  demands, resource prices
• We want to compare conventional approach with
  BDIM approach
• Conventional approach uses ad hoc values
• Tmax 1.5 seconds, Amin 99.96

16
Cost comparison

• Conventional seems better (cheaper) yearly cost
  324,600 vs 345,400

17
But let's look at provider profit

• Profit 12 better, on average

18
Is the client doing badly because of better
provider profit?

• Service costs increase but profit increases even
  more!
• Client spends 38,550 more with the
  business-driven approach while raising profits by
  almost 562,630

19
SLA design

• SLOs are simply calculated from optimal
  business-oriented design
• Tmax 0.112s and Amin 99.968

20
conclusions
21
Summary

• We have presented a business-driven method to
  design SLAs for outsourced IT services
• Approach is to link provider and client financial
  results and tie these to the infrastructure
• Instead of defining SLA parameters in an ad hoc
  way, we optimize the business results and simply
  pick off the SLA values

22
Conclusions

• We have shown that SLA design from a business
  perspective can
• Be used to choose SLOs, rather than uding the old
  finger in the air approach
• Yield better financial results for both service
  client and provider businesses
• An optimal tradeoff can be found that is
  beneficial to both

23
Future Work

• Major weakness is that business-IT linkage model
  is only applicable to an e-commerce type scenario
• Need to investigate linkage models for other
  types of business processes
• Need to investigate other dimensions in the SLA
  design space such as how to define optimal
  penalty and reward functions
• Rethink SLAs to put business-oriented SLOs?

24
Thank you!