Project Status

advertisement
Robustness
849 Dependable Embedded System
-- Ying Shi
Mar.4.99
Outline
QoS
Robustness
Control Theory
Exception Handling
Manufacturing/
Quality
 Basic concepts
 Metric -- RB , major of this talk
 other Robustness aspects + topic connections
 Conclusion
 Reference List
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-2
What’s Robustness
 One of those system attributes such as
performance
 IEEE Standard Glossary
– “The degree to which a system or component can
function correctly in the presence of invalid inputs or
stressful environmental conditions.” error tolerance,
fault tolerance
 specifically might be interpreted as the degree
of the system ability of
– handling exceptions
– tolerating faults ...
 One of the concerns -- establish Metric for this
attribute
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-3
Robustness Benchmark (RB)
 An identification of system error detection
mechanism(EDM)
 A RB should measure how a system reacts to
possible erroneous inputs or environmental
factors that could affect the system and result
in failures.
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-4
System component Hierarchy
 Robustness measure of a combined
components system would provide an
indication of how erroneous or unexpected
inputs are handles by all levels of abstraction
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-5
Cover both HW & SW
 Both HW errors and SW errors to be covered
 Many HW EDMs are embedded in the design
(CPU, mem, and I/O subsystems). Errors
detected at this level, reflected to the SW level
primarily through traps, almost always
gracefully accommodated by the error handler
(except crashes)
 SW level, ideally, EDMs should be embedded
within each of these levels to detect, isolate,
and accommodate(via error handler)errors
from coding or possibly design.
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-6
Reality Restricts
 Ideally, to exercise all possible levels of error
detection, trapped by both HW and SW, w/ the
end result being a measurement w/ respect to
an absolutely robust system
 In reality,
– exhaustive testing is not possible
– natural errors occur infrequently enough that simply
observing system operation is too time consuming.
 Thus RB would serve as a relative measure of
robustness or fault tolerance between two
systems/machines.
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-7
Living with the Restriction
 Try to cover more and make reasonable
combining
 Measure different system components::
–
–
–
–
input error checking capability of system software
file system management
utilities
library routines ...
 quantitatively, how much of each metric weighs
towards the final overall system robustness
figure
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-8
More of Robustness
 Techniques towards Robustness
–
–
–
–
–
classical HW FTC techniques
graceful degradation
fail-safe
idiot-proof (UI)
functional diversity
– etc...
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-9
Connections
 Exception Handling
– one of the means towards robustness
 Control Theory
– nah...leave with leo’s later talk
 Quality of Service
– robust system have the ability to guarantee QoS
– they are two individual system attributes
 Manufacturing/quality
– robustness is the drive during manufacturing
– robustness - one of system qualities
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-10
Conclusion
 Robustness refers to one of the system
attributes, also a goal in design
 Complexity makes it impossible to have an
ideal RB, thus as a relative metric b/w systems
 RB as a guide for design
 Robustness is quite a broad topic, I assume,
there is too much to dig in the future to say any
specific aspects of it are the main future trend
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-11
Reference List
 Required paper :: Development of a Benchmark
to Measure System Robustness, Siewiorek,
Hudak, Suh, Segall, 23rd FTCS
– insight into developing a system level RB
– several primitive benchmarks, and combined into RB
suites
– a methodology to combine the primitive RBs to form
an overall robustness figure
 Journal
– FTCS; R&M proceeding
Robustness - Dependable Embedded System
Ying Shi
Mar.4,99
-12
Download