Designing User Interfaces Spring 1999

1
SE 767-NT Software Performance Engineering Robert
Oshana Lecture 31 For more
information, please contact NTU Tape
Orders NTU Media Services [phone]
oshana_at_airmail.net
tapeorders_at_ntu.edu
2
Where we are

• Introduction
• SPE Quick View
• SPE and the UML
• Software Execution Models
• Web applications and other distributed systems
• System execution models
• SPE data collection
• Software measurement and instrumentation
• Performance oriented design
• Performance patterns
• Performance anti-patterns
• Implementation solutions
• Web applications
• Embedded and real-time systems
• The SPE process
• Implementing SPE

3
Implementing SPE

• Chapter 16

4
Introduction

• You can personally start using SPE right away
• Trying to do this while working in a group can be
  difficult
• This process has huge benefits, but like other
  technology insertion, it cannot happen overnight

5
Tools

• Modeling tools
• System modeling tools
• Most support the queue-ing network model paradigm
  (mostly capacity planning)
• DIGITAL Capacity Planner (Compaq - HP)
• Software modeling tools
• Only a few tools specifically for SPE
• SPE-ED
• Since these are solved analytically, other tools
  can be customized for this process

6
Tools

• Development tools
• No specific tools for SPE
• Can customize existing tools

7
SPE adoption and use

• SPE has been adopted with success at several
  companies (Lucent, IBM, Yantra, Aprisma
  Management)
• Several factors to consider
• Organizational
• Managerial
• Technical

8
Establishing a SPE practice

• starts small, and, once organization recognizes
  value, it becomes an established part of the
  development process..
• Obtaining commitment from the projects is
  important (not always easy)
• Start small and build on a string of successes

9
Establishing a SPE practice

• You will never convince everyone
• Start with small projects
• Concentrate on new parts of the application as
  they evolve

10
Qualifications

• Not just a modeler but also a good communicator
  willing to bring bad news into the open
• Get training to increase credibility on projects
• Offer training to other groups

11
Other advice

• Build and measure prototypes early in development
• Apply centering principle and fast path to
  critical use cases and it is hard to go wrong
• Start getting performance goals early
• Do performance walkthroughs

12
Other advice

• Publicize success
• Actively participate in professional societies
• Workshop on Software and Performance (WOSP)
• Publish

13
Key considerations

• Economic justification for SPE
• Responsibility for SPE
• Finding the right people
• Proper organizational placement

14
Pilot projects

• A pilot project is a small-scale project that is
  conducted to test a process under realistic
  conditions
• Manageable
• Non-trivial
• Non-critical
• Measurable

15
Critical success factors

• Find a string advocate
• Secure commitment
• Integrate SPE with development process and
  schedule
• Develop proper tooling
• Secure cooperation towards goals
• Establish credibility

16
Implementation strategies

• Centralized SPE group
• Performance engineers as part of the development
  team
• A combined approach

17
Critical factors for successful projects

• Establish performance objectives and hold
  managers and developers accountable
• Produce credible model results
• Get consensus on set of performance scenarios
• Measure critical components early and often
• Use best and worst case results

18
SPE future

• Will we ever have enough computing power to not
  worry about SPE ?
• Hasnt happened yet !!
• Portable embedded systems cannot use more power
  as the solution to all performance problems
• Limitations include tools, process adoption, time

19
SPE Summary
Existing work
SW execution model
System execution models
Performance metrics
At 90 utilization, costs double, at 95
utilization they triple!
Resource requirements

• Workload worst case scenarios
• Performance objectives quantitative criteria for
  evaluating performance (CPU, memory, I/O)
• Software characteristics processing steps for
  various performance scenarios (ADD, simulation)
• Execution environment platform on which the
  proposed system will execute, partitioning
  decisions
• Resource requirements estimate of the amount of
  service required for key components of the system
• Processing overhead benchmarking, simulation,
  prototyping for key scenarios

HW Configuration
Workload
New SW
Capacity
20
SE 767-NT Software Performance Engineering Robert
Oshana End of lecture For
more information, please contact NTU Tape
Orders NTU Media Services [phone]
oshana_at_airmail.net
tapeorders_at_ntu.edu