Software Runtime Adaptation: Framework, Approaches, Styles
Peyman Oreizy
Launch21
Nenad Medvidovic USC
Richard N. Taylor UC Irvine
What? Our paper got the MIP award?!
2nd thought: Wow, lots of citations! Original paper: 315 Follow-on journal paper: 375 690 papers over ~9 yrs 1.5pppw 3rd thought: Could one person be responsible for all of them?
1st thought: Someone read our paper!
Change during runtime?
Critical systems require "continuous availability"
Power grid, financial systems, ...
Increasingly important in everyday systems
How did we get here?
Serendipity Key insights:
connectors explicit arch-model fielded with the system and used to govern change architectural style
State of the Practice
"hot pluggable" drives and memory system virtualization (ala VMware and Xen) programming language facilities for dynamic loading, linking, and patching
of code binary code patching redundant and fault-tolerant hardware
software designed for fault tolerance
State of the Practice
Each approach has its place
No one approach encompasses the others Clear benefits to enacting change at multiple levels of abstraction
Need a framework for comparing and combining approaches
Towards a Unifying Framework
1.Use a "model" to highlight some
system details while hiding others
2.Grapple with 5 aspects:
a. evolve behavior b. evolve state c. adjust execution context d. asynchronous change
A Look Back
What has happened in the past decade?
Research projects Open-source and commercial systems
Conferences, symposia, and workshops
Dynamic adaptation models
Dynamic Adaptation Models I
Prior to our ICSE 1998 paper
Style-based models: CHAM, graphgrammars ADL-based models: Darwin, Dynamic Wright, Rapide
Did not gain wide adoption
Constrained notion of dynamism
Lack of system-level facilities
Dynamic Adaptation Models II Subsequent to our ICSE 1998
paper
"Figure 8" model: system adaptation driven by architecture
Dynamic Adaptation Models III
Self-managed dynamic systems: plan generation
Rainbow: similar to "Figure 8"
Research Projects
MobiPads: QoS optimization via dynamic reconfiguration
Siena: Client-, server-, and network-level dynamism Grid computing: Dynamic addition and removal of computing resources Aura: QoS-driven system reconfiguration
Commercial Solutions
Koala: predefined dynamic adaptations via options Skype
P2P-based adaptations
Promotion/demotion of nodes
MapReduce: automatic data rerouting from failed to live nodes
Conferences/Symposi a
Dynamism as a means or by-product Dynamism in flagship SE conferences
Dynamism as primary focus
Dynamism as Primary Focus ICAC ACW CHIAACS SEAMS WOSS DEAS WADS IWPSE Dagstuhl SESAS
Dynamism as By-Product
PerCom CD Middleware
MobiCom
Dynamism in SE Conferences at: What happened to dynamism
FSE ASE
ICSE
What about software architecture venues: WICSA
ECSA QoSA CBSE
Promising Directions
A simple message: if you want or need adaptable applications you can either:
Make no constraints on developers
... and then work like crazy to try to obtain adaptation
Constrain development to make adaptation easier and predictable
This should not be news: the message is s...