A Process Model of Developing Micro-code for a Network Processor

1
A Process Model of Developing Micro-code for a
Network Processor

• 18.10.2005 Jani Koski

Author Jani Koski Supervisor Prof. Raimo
Kantola Instructor M.Sc Marko Lohjala
2
Contents

• Introduction to the study
• Network device
• Network Processor (NP)
• Micro code for NP
• Why research was made
• Research problem
• How to make micro code?
• Micro code SW process?
• Research method
• Search best SW model from literature
• Lets take to account the nature of Micro code,
  environment, etc
• gt case study
• Case study
• Project analysis
• NP analysis
• Micro code analysis
• The result

3
Network device
Network Processor (NP)
4
Network Processor (NP)
Micro code
5
Why research was made

• Previously micro code was done by Agere Systems
• But also the Nokia had knowledge to produce the
  micro code
• Faster SW integration if Nokia produce the micro
  code it selves?
• Nokia did not have any process for producing
  APP550 specific micro code

6
Research problem

• In what level micro code SW process should be
  investigated?
• Schedule of micro code SW process? When to
  implement micro code?
• How to implement micro code to support HW-test?
• The nature of micro code? How does it differ from
  traditional coding?
• How do we test micro code? What tools do we have?
• Can we reuse some generic SW process models using
  literature?

7
Answers (Q1)

• In what level should question micro code SW
  process? be investigated?
• The level is Generic micro code SW process model.
• It means that we will find different generic SW
  process models (e.g. Spiral model, Water Fall
  model) from literature and approach our question
  from academic point of view.
• We avoid copy-paste method and existing Nokia
  SW process models are not used
• To choose best model from literature we need
  information about nature of Micro code.

8
Answers (Q2)

• Schedule of micro code SW process? When to
  implement micro code? Somewhere middle of HW and
  SW schedule?
• HW must get some preliminary version from micro
  code to make their HW tests (HW MT). Without
  micro code it is not possible.
• Schedule is most likely following

Final version
Test version to support HW test
simulation
Test board (e.g. Hydra)
9
Answers (Q2) cont.

• How to map the micro code SW process to our HW
  milestone?
• We have different kind of SW to be integrated
• OS, SW, Micro code
• For micro code at least four different versions
  can be identified
• Preliminary micro code version 1 test with
  simulator
• Preliminary micro code version 2 test with test
  board
• Preliminary micro code version 3 to support HW
  test
• Final micro code version 4 with all feature
  support

Final version
Test version to support HW test
simulation
Test board (e.g. Hydra)
10
Answers (Q3)

• How to implement micro code to support HW-test?
• We need some input from HW group to make
  sufficient first version from micro code to HW
  group.
• After knowing requirements we can implements
  micro code.

11
Answers (Q4)

• The nature of Micro code? How does it differ from
  traditional coding (like C, C, Java) or does it
  differ at all?
• Yes, micro code differs from traditional coding.
  If we compare e.g. to C language and SW done with
  that, micro code is
• Usually less code lines than in C
• gt no big resources needed
• Available memory is usually small
• Big micro code software not possible
• Available memory can not be used as in C (we have
  no pointers).
• Instruction set smaller than in C
• Same code editors can be used (emacs, UltraEdit)
• Micro code can be simulated with simulator (SPA
  if APP is used)
• Note that there are different versions of APP and
  simulators!
• Debugging is different (and difficult) in micro
  code (printf(Hello world) is not possible).

12
Answers (Q4) cont.

• The nature of Micro code? How does it differ from
  traditional coding (like C, C, Java) or does it
  differ at all?
• Yes, micro code differs from traditional coding.
  If we compare e.g. to C language and SW done with
  that, micro code is (APP)
• Real time programming
• In micro code we have only very limited time to
  use per received cell or PDU
• In micro code we may have dynamic changing code
  (using searching trees). Trees are usually
  modified by some host processor.
• Micro code compiler gives similar kind of warning
  and error than e.g. gcc.

13
Answers (Q5)

• How do we test micro code? What tools do we have?
• HW can not be ready if it is not verified by HW
  MT (HardWare Module Testing). HW MT needs some
  preliminary micro code for testing. But how do we
  create micro code if we can not test that in some
  HW?
• Chicken-Egg problem
• Because we do not have HW ready for testing we
  need simulator. Usually manufacturer can offer a
  simulator for their NP (e.g. SPA for APP). But
  simulator may have some SW bugs and we can not be
  100 sure that micro code is really working.
• We could use also test boards that include the
  real NP (delivered by manufacturer). E.g Ageres
  Hydra.
• When our HW is ready we will continue testing
  with that.
• Testing equipment needed! (e.g. packet generator
  etc.)
• Simulator can be still used.

14
Answers (Q6)

• Can we reuse some generic SW process models using
  literature?
• Generic SW process models from literature
• Other SW process models.
• Agility methods in SW process models.

15
Choosing the SW process

• Generic SW process models from literature
• Water fall model is too heavy process model for
  micro code. It produce lot of documentation and
  its applicable for big process
• Evolution model is a good model for producing
  complex SW. SW is produced iterative but that
  will cause lot of updates to documentation. The
  biggest disadvantage is poor SW architecture
  caused by this model.
• Component model uses old SW as much as possible.
  This leads SW design where SW components are very
  generic. However, the result is usually big SW
  because code is not optimized.

16
Choosing the SW process

• Other SW process models
• Prototyping is very useful model for example
  finding out critical capacity questions
• RUP (Rational Unified Process) is a combination
  of Water fall, Evolution Component models
• Formal models is based on mathematical forms
• 4 GT models are higher level tools for generating
  code
• Code and fix model is a poor model
• Spiral model is a model for minimizing risks

17
Choosing the SW process

• Agility methods in SW process models
• XP (Extreme Programming) was used as a main model
  for micro code
• RAD (Rapid Application Development) is developed
  from 4 GT models

18
XP-model

• Based on iteration
• Pair programming
• For small projects (e.g. about 2-12 programmers)
• Promotes team work
• Stresses customer satisfaction (e.g. other
  project)
• Customer is part of the process
• Developers, managers and customers are working
  together
• Design unit test case before coding
• Automatic testing

19
The Micro code SW process
Host processor interface specification
documentation DB
20
Detailed micro code SW process
21
Conclusion

• Feature IP QoS was implemented by using the micro
  code SW process
• Micro code SW process works but is it efficient?
• Requires measurements later (out of this thesis)
• Component model was used but XP model is
  dominating
• Pair programming in XP model may face some
  problems. E.g cultural and behavioural problems

22
Questions?
Comments?
Any new ideas?
Thank you!