architectural mismatch or why it’s hard to build systems out of existing parts

23
Architectural Mismatch or Why it’s hard to build systems out of existing parts

Upload: george-burke

Post on 23-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Architectural Mismatch or

Why it’s hard to build systems out of existing parts

Bibliography• Garlan, Allen, Ockerbloom: Architectural Mismatch or

Why it’s hard to build systems out of existing parts, in ICSE 1995

– Article has been nominated as one of the “most influential” in a 25-years retrospective of IEEE Software and had a follow-up:

• Garlan, Allen, Ockerbloom: Architectural Mismatch: Why Reuse is Still So Hard, IEEE Software 2009

• Ivica Crnkovic, Magnus Larsson. Building reliable component based software systems,  Chapter 9: Component Composition and Integration

Review: Component Integration

• Integrating components can be illustrated as a mechanical process of “wiring” components together to form assemblies (connecting required and provided interfaces)

• Standardization in form of component models reduces some of the integration problems

Review: Component Composition

• Integrating (“wiring”) components is only a first step

• Composition (making components “play well together”) is difficult:– We have seen that interface compatibility (both syntactic

and semantic) is needed

Component Composition Problems

• the problem goes beyond interface compatibility (syntactic and semantic) and could be due to the so called Architectural mismatch

Architectural mismatch

“Architectural mismatch stems from mismatched assumptions a reusable part makes about the structure of the system it is to be part of. These assumptions often conflict with the assumptions of other parts and are almost always implicit, making them extremely difficult to analyze before building the system.”

D. Garlan, R. Allen and J. Ockerbloom. “Architectural Mismatch: Why Reuse is So Hard,” IEEE Software, 12(6):17-26, November 1995

http://www.cs.cmu.edu/afs/cs/project/able/ftp/archmismatch-icse17/archmismatch-icse17.pdf

Component composition issues

• “Architectural mismatch stems from mismatched assumptions a reusable part makes about the structure of the system it is to be part of”

• four classes of structural assumptions – The nature of components (infrastructure, control model, and data model)

– The nature of connectors (protocols and data models)

– The architecture of the assemblies (constraints on interactions)

– The run-time construction process (order of instantiations).

Architectural mismatch examples

• 2 case studies:• D. Garlan, R. Allen and J. Ockerbloom “Architectural

Mismatch: Why Reuse is So Hard”

• AESOP

• P. Inverardi, A.L. Wolf, and D. Yankelevich, Static Checking of System Behaviors Using Derived Component Assumptions

• Compressing proxy

Compressing proxy example

• Problem: adding data compression to a web server, in order to improve performance

• Characteristics of web server: – Standard CERN HTTP server– Pipes-and-filters architecture

• Proposed solution:– Add an external filter to compress data using gzip– External compressing filter communicates with the web server

through Unix pipes (read and write data)– A pseudo filter (adaptor) is added to work with the external

compressing filter

Compressing proxy example

Filter FilterPseudo Filter

(Adaptor)

gzip

1

2 3

4

Compressing Proxy

Process

Component

Channel

Function call interface

UNIX pipe interface

Figure 9.1 from [CrnkovicBook]

Compressing proxy issues

• HTTP server filters are forced to read when data are pushed at them

• Unix filters choose when to read data -> gzip may block• Normal scenario:

– Adaptor passes the data from input filter to gzip, and when the stream is closed, reads the data from gzip and writes in the output filter

• Problematic scenario:– If the input file is large, because gzip has a limited internal buffer,

it may attempt to write a portion of compressed data before the adaptor is ready: => deadlock (gzip blocks and adapter is blocked)

• Solution:– Adaptor should handle data incrementally and use unblocking

read and write

Lessons learned from Compressing Proxy

• Formal architectural description and analysis to uncover what they call “behavioral mismatch”

• Not a component mismatch (components can be “wired” together)

• Components must express behavioural properties:– assumptions made about it’s environment such as data formats or

buffer sizes

– Its effects on the environment

AESOP Example

• AESOP: developed by CMU, an environment that generates development environments tailored for systems of a particular style

• Development approach: integrate existing components• Components:

– A object oriented database (OBST – public domain system)– GUI toolkit (Interviews and Unidraw – Stanford university) – An event-based tool-integration mechanism (SoftBench from

HP)– A RPC mechanism (Mach RPC Interface Generator – CMU)

• Advantages:– Stable tools, used in several other projects– All tools implemented in C/C++ AND with source code available

Problems in AESOP integration (1)

• Assumptions about the components: infrastructure:– Components assume they have a certain infrastructure, but it is not available

• Ex: SoftBench assumes all components have their own GUI interfaces and hence have the X-library’s communication primitives loaded => all components had to load X library even if they do not need it otherwise => code bloat

– Components assume that they should provide a certain infrastructure, which the application does not need

• Ex: OBST provides many functions, Aesop needs only few of these

Problems in AESOP integration (2)

• Assumptions about the components: control model:– Different packages make different assumptions about which part

holds the main thread of control• Ex: SoftBench, InterViews and MIG all use event loops that are

incompatible with each other => rewrite InterViews event loop

Problems in AESOP integration (3)

• Assumptions about the components: data model:– Different components assume different things about the nature

of data• Ex: Unidraw maintain a hierarchichal model for its objects but allows

only top-level objects to be manipulated by users, Aesop requires that both parent and child objects can be manipulated => rewrite Unidraw hierarchy

Problems in AESOP integration (4)

• Assumptions about the connectors: protocols

• SoftBench handles RPC by mapping it within the event framework through 2 events (the request and the reply) => the caller becomes more complicated => use Mach RPC instead of Softbench

Problems in AESOP integration (5)

• Assumptions about the connectors: data model– Components have different assumptions on what comes

over a connection• Ex: SoftBench: Strings; MIG: C data; OBST C++ data.

• translation between formats needed => performance bottleneck

Problems in AESOP integration (6)

• Assumptions about the global architecture– OBST assumes that all communications occurs in a star

configuration, with itself in the center, Assumes independence of client tools and provides a transaction protocol per single tool, not per combination of tools

– Aesop’s communication structure is a more general graph (tools cooperate also directly) => OBST’s transaction mechanism can result in deadlock or database inconsistency => write own transaction mechanism

Problems in AESOP integration (7)

• Assumptions about the building process– Assumptions about the library infrastructure

– Assumptions about a generic language (C++)

– Assumptions about a tool specific language• Some component A may have other expectations on the

generated code of another component B as B itself

Consequences for Aesop• Effort:

– Estimated: one person-year– Reality: 5 person-years

• Code bloat• Poor performance: due to overhead of the tool-to-database

communication and excessive code• Need to modify existing packages: ex. Event loop of SoftBench and

Interview• Need to reimplement some existing functions: OBST transaction

mechanism, Hierarchical nested view management in InterView

Component integration issues

• “Architectural mismatch stems from mismatched assumptions a reusable part makes about the structure of the system it is to be part of”

• four classes of structural assumptions – The nature of components (infrastructure, control model, and data model)

– The nature of connectors (protocols and data models)

– The architecture of the assemblies (constraints on interactions)

– The run-time construction process (order of instantiations).

Recommended practice

• To avoid architectural mismatch:– Make assumptions explicit

– Use orthogonal loosely-coupled components

– Use bridging techniques to solve mismatches

– Provide design and composition guidelines