Engineering Challenges of Embedded Software

1
Engineering Challenges of Embedded Software

• John LinnDirector, Systems and Software
  LaboratoryTexas Instruments Inc.

2
Texas Instruments (TI) Importance of Embedded
Software

• TI is a major semiconductor manufacturer
• Digital Signal Processing (DSP) integrated
  circuits
• The analog components that transform real-world
  signals to digital
• Most of our products are programmable
• Enable products with embedded software
• Programmed by our customers (and us too!)

3
Agenda

• Embedded software What is it? What makes it
  different?
• Computer economics and how financial decisions
  affects embedded software engineering
• Schedule and the challenges of quick
  time-to-market
• Discussion and QA

4
What is Embedded Software
5
What is Embedded Software
Some Characteristics

• Low cost
• Close to the hardware
• Software on a special-purpose processor
• Very high performance
• Low power and power management
• High reliability
• Real-time
• Small footprint
• Software/firmware in ROM/PROM

6
Embedded SoftwareMy Definition

• Embedded software is software that is ultimately
  bundled with hardware and sold to the end-user as
  a complete product
• It usually has some (or all) of the
  characteristics on the previous slide

7
Some Products ContainingEmbedded Software
8
Close to the HardwareDigital Still Camera

• Characteristics
• Multiprocessing
• Different ISAs
• ASIC custom processor
• Complex low-level peripherals
• Power management
• Tight memory

9
Low-CostMobile Handset

• Characteristics
• Very low cost
• Multiprocessing
• Different ISAs
• Very complex low-level peripherals
• Complex power management
• Tight memory
• Often ASIC custom processor
• Real-time deadlines
• High reliability

10
Computer Economics

• The cost of software
• Recurring vs. non-recurring costs
• Software reuse and the value web

11
Increasing Cost of Software Development

• Software development costs are going up while
  hardware costs are decreasing whats going on?
  The CEO

12
Memory Prices Drives Software Content
13
Memory Prices Drives Software Content

• Our Customers
• I can afford to put 50 Mbytes of memory and 200
  Mbytes of ROM in my product. I want to fill it
  up with quality software features that sell

14
Software Contribution to Product Cost

• Embedded software has become a significant
  contributor to product cost
• Growth in demand for functionality results in
  exponential growth in complexity (lines-of-code)
  in products
• Increased complexity more than offsets improved
  programmer productivity

15
Computer EconomicsUnit Cost vs. Production Volume
Production Cost per Unit
Number Units Produced per Model
16
Computer EconomicsRecurring and Non-Recurring
Costs

• Non-Recurring Cost Incurred once
• Specification
• Design
• Coding
• Testing
• Integration
• Recurring Cost Incurred for every item produced
• Manufacturing material and labor
• Some types of royalties
• Manufacturing engineering
• Waste, yield, quality defects
• Maintenance, repair

17
Computer EconomicsRecurring and Non-Recurring
Costs

• Hardware production is almost exclusively
  recurring cost
• Software development is almost exclusively
  non-recurring cost
• This one-time cost must be amortized across all
  the products sold
• Significant reduces software cost by spreading
  over larger number of products

18
Computer EconomicsUnit Cost vs. Production Volume
Production Cost per Unit
Number Units Produced per Model
19
Computer EconomicsAmortization of Non-Recurring
Software Cost

• F/A-22
• 3M LOC _at_ 3 LOC per day 600M
• 6-7 years
• Amortized cost 4M per production aircraft (1
  of recurring cost)
• Competitors Now there are none
• If a competitor could cut that cost in half,
  would it matter? - Not much

20
Computer EconomicsAmortization of Non-Recurring
Software Cost

• F/A-22
• 3M LOC _at_ 3 LOC per day 600M
• 6-7 years
• Amortized cost 4M per production aircraft (1
  of recurring cost)
• Competitors Now there are none
• If a competitor could cut that cost in half,
  would it matter? - Not much
• 2.5 G Cellular Handset
• 2M LOC _at_ 10 LOC per day 120M
• Amortized cost 12 per production phone (25 of
  recurring cost)
• Competitors Everybody
• If a competitor could cut that cost in half,
  would it make a difference? Big cost advantage!

21
How Do Our Customers Survive?
22
How Do Our Customers Survive?Software Re-Use and
the Software Value-Web

• Value-web of software customers and vendors
• Significant software re-use across multiple
  products and customers
• Resulting lower amortized cost of software allows
  focus on recurring costs
• Interaction between software and hardware really
  important for cost

23
Cost of Embedded Software Key Points

• Learn when you need re-usable software and when
  you dont
• Look at functionality and recurring/non-recurring
  costs as a guide
• Learn how to make software re-usable
• Your customers will also be software developers
• Learn how to use re-usable software from other
  vendors
• You will also be a software customer

24
Computer Economics

• Value of software development predictability

25
Computer Economics Predictability of Software
Level of Effort

• Observations
• Predictability business success
• This is why CMM places such emphasis on
  predictability
• Biased estimates (over optimistic, over
  pessimistic) make it even worse
• Pricing strategies can help avoid the big
  time losses, but cant totally compensate

26
Computer EconomicsValue of Software LOE
Predictability

• Assumptions
• Head-to-head bidding competition
• Same software productivity and quality at both
  companies
• Company A costs are 2x less predictable than
  company B costs

Adjusting pricing wont make up for lack of
predictability
27
Computer EconomicsCosts of Software
Predictability

• Observations
• You cant afford too much loss in efficiency to
  achieve predictability
• The big argument against heavyweight
  methodologies

28
Computer EconomicsLevel-Of-Effort Prediction
Models

• You usually need a statistical model to predict
  stuff
• Its a statistical world - random things happen
• Even if it were deterministic, software
  development is too complicated to anticipate in
  sufficient detail
• You need some factors (parameters) that correlate
  to the final level of effort to base your model
  on
• Software LOE models require
• Fundamental mathematical description of cause and
  effect
• Input parameters like lines of code, code
  complexity, team expertise, etc.
• Mid-course correction, if available, during the
  project
• Calibration
• Improvements over time for better precision
• Human assessment, judgment and adjustment

29
Computer EconomicsParametric Models
Model Quality
Parameter Estimation Quality
30
Computer EconomicsCOCOMO

• Barry Boehms Constructive Cost Model
• Lines of Code used as a complexity estimate
• Lots and lots of other project and team
  parameters
• Significant industry calibration
• Big argument over whether LOC is a good metric
• See USC Center for Software Engineering
  http//sunset.usc.edu/
• If you want to play with it http//sunset.usc.edu
  /research/COCOMOII/

31
Computer EconomicsFunction Points

• Allan J. Albrechts Function Point Analysis
• Function point count used as measure of
  complexity
• Transaction-based approach looking at amount and
  complexity of control and data crossing component
  boundaries
• Replaced LOC with (presumed) more representative
  complexity measure, but uses other aspects of
  COCOMO for conversion to LOE
• See http//www.ifpug.com/fpafund.htm

32
Software Cost PredictabilityKey Points

• Development level-of-effort predictability is
  critical
• You need discipline (methodology)
• You probably need a model to help predict cost
• Learn to use one, but temper it with judgment
• You need to keep metrics to calibrate it
• You can improve predictability with process
• But you cant afford to overdo it

33
Computer EconomicsSchedule

• Time is money
• - Ben Franklin

34
Time is money, professor proves May 29, 2002
Posted 1230 PM EDT (1630 GMT)

• WARWICK, England (CNN) -- A mathematical formula
  calculated by a British university professor has
  found that time actually is money.
• According to the equation, the average British
  minute is worth just over 10 pence (15 cents) to
  men and eight pence (12 cents) to women.

The formula is V(W((100-t)/100))/C, where V is
the value of an hour, W is a person's hourly
wage, t is the tax rate and C is the local cost
of living. It shows that there is no such thing
as a free lunch or even a free dinner, while
brushing your teeth for three minutes uses up 30
pence (45 cents) in "lost" time, and washing a
car by hand has a hidden cost of 3 (4.50).
35
Schedule Importance of Schedule and
Time-to-Market

• For products with 9-month development-cycles,
  time-to-market is everything
• Consumer products typically start right after
  Christmas and go to production for the next
  Christmas season (Aug. or Sept.)
• Handsets go to production when service providers
  roll out new services
• First to volume sales last to remain profitable
• Compressed schedules and co-design are common
• HW/SW simulation development and test
• Spartan test and debug environment and limited
  test hardware
• Cooperation between HW/SW teams in different
  companies/countries
• Dynamically changing (read chaotic) requirements
• The development methodology required may be
  different than the waterfall model used in
  (F/A-22 like) 10-year projects
• Time-to-market often trumps efficiency
• Shortened schedules are usually inefficient

36
ScheduleHardware/Software Co-Design

• What do you do when the integrated circuit and
  all the peripherals are not ready?
• What do you do when the other software in the
  value web is not yet compete?

37
ScheduleHardware/Software Co-Design

• Virtio Software Emulator
• Runs actual TI OMAP1510 Platform targeted
  binaries
• Includes real-world connections such as serial,
  network and video streaming
• Provides high level of system visibility by
  viewing of system registers and variables inside
  peripheral hardware models

• APTIX Hardware Prototyping System
• Verify complex system-on-chip designs
• Proprietary reconfigurable prototyping methodology

38
Schedule Software Hardware Co-Design Simulation
/Emulation Headaches

• Speed
• 100x - 10,000x slower than real time
• Fewer, longer tests
• Timing
• Often cannot reproduce accurate timing sequences
• Difficult to find race, deadlock and performance
  issues for real time
• Accuracy
• Simulation models are software and have bugs, too
• Significant development effort for verification
  and diagnostics
• Cost
• Up to several millions of dollars for
  timing-accurate hardware simulators
• Development environment
• Usually very Spartan
• Few tools are ready when you need them and
  whats there is buggy
• Other parallel software efforts complicate
  things, too
• Device drivers, OS ports, debug tools, etc.

39
ScheduleProduct Testing Headaches

• Software test requirements often not considered
  in product design, such as
• Rapid software download/install
• Bulk-data transfer
• Reliable communications
• Hardware diagnostics
• Run/reboot buggy software automated and
  unattended
• Special test framework and harness
• Test automation host

40
ScheduleKey Points

• Schedule is as important as cost
• Learn how to trade cost for schedule
• (See re-useable software)
• Short schedules cause development headaches
• Get your development requirements included up
  front by the hardware developers
• Create a test integration plan that takes
  hardware limitations into account
• Buy/build enough emulators/simulators/prototypes
  to execute your plan
• Plan for problems with the hardware and other
  vendor software

41
Conclusions

• Embedded software is now big, complex business -
  extreme discipline and organization is required
• Economics determines what is important - products
  have different economies and we need our software
  engineers to understand and be flexible
• Our customers (and we too!) need more emphasis on
  software engineering for embedded software - its
  where the action is in todays economy

42
What Do You Need to Do as Software Engineers?

• You need to understand the economics (business
  case) of the system your software is used in
• You need to be organized and use an appropriate
  methodology
• You need to understand your organizations value
  web and how to make software re-usable
• You need to know when to re-use and when to
  create from scratch
• You need to be flexible and agile One
  methodology does not fit all situations
• You need to be skilled and experienced