Brief Introduction to CosMIC:
The
Component Synthesis using Model Integrated Computing (CoSMIC)
project is a MDA toolset being developed by the Institute for Software
Integrated Systems (ISIS) at
1. Model and analyze distributed real-time and embedded application functionality and QoS requirements.
2. Synthesize CCM-specific deployment metadata required to deliver end-to-end QoS to DRE applications.
The CoSMIC toolsuite provides modeling of DRE systems, their QoS requirements, and QoS adaptation policies used for DRE application QoS management. The component behavior, their interactions, and QoS requirements are modeled using a domain specific modeling language that adheres to the OMG Deployment & Configuration (D&C) specification (ptc/2003-07-08). CoSMIC enables modeling the standards-based CCM components.
Hand-writing the required XML descriptors to configure our components is tedious and error prone. We therefore utilize CosMIC (we specifically use a subset of CosMIC called PICML, which stands for Platform-Independent Component Modeling Language) to generate the bulk of descriptors for us.
For more information on CosMIC, please refer to the following article:
CoSMIC:
An MDA Generative Tool for Distributed Real-time and Embedded Applications.
Importing IDL to PICML
To quick start our Stock Quoter modeling
process in GME, CoSMIC introduces idl_to_picml , which is an
executable program that imports the IDL files you just created into PICML.
%COSMIC_ROOT%\bin
is included in
the PATH variable, e.g., c:\Program Files\ISIS\CoSMIC\bin.MyQuoter
\
, and type the following command:MyQuoter.xme
file in the MyQuoter
directory. Note
that if -x option is not used, the default xme
file name will be PICML_default_xme_file.xme
.
Figure 2
In the Browser of the generated model, the PredefinedTypes folder contains a bunch of atomic datatype elements that will be referenced by other modeling
parts later. You don't have to worry about them for now. The models in InterfaceDefinitions folder are the PICML
representations ofthe IDL files you just imported.
Double-click to open Broker, it will show you a white, document-like
entity which represents a <<FileRef>>
instance and a yellow box-shaped entity representing a <<Package>>
instance. Their equivalent representations in IDL files are
"#include" and "module" respectively. Figure 3 gives you a
clear view of GME representation of all the 3 idl
files created. We will cover the other folders in the next section.
Figure 3
Now
we are ready to model the rest of the Stock Quoter
systems.
This section describes modeling the Quoter application using PICML. If you have trouble
producing a functioning model from this tutorial, please see the, pre-built Quoter model
which is provided for your reference. This model contains all elements created
as part of this tutorial.
Note: To import an XML file in GME, select File->Import
XML.. from GME and choose your XML model
The PICML paradigm is designed for the OMG Deployment &
Configuration (D&C) specification (ptc/2003-07-08), so the modeling
process is straightforward if you are familiar with the specification. Please
see the DAnCE
project for more information. For those who hate specification
(including me, :-)), an overview of D&C
and as well as a tutorial of
D&C and CCM can help to reduce the learning curve. (Thanks to Sowayan, Abdulah for pointing this out). In addition, it is also
helpful to read through Chapter 32 of TAO's
Developer's Guide 1.4a (CIAO and CCM) which thoroughly describes the various
descriptors required for a component (*.iad, *.ccd, *.cid, etc).
The complete PICML Quoter
model consists of modeling elements distributed across various folders. If you
used idl_to_picml to generate the initial
model, you will see that it has created all these folders and some of the
tedious boilerplate modeling for us. The rest of the section will explain the purpose
of each folder as well as the modeling entities contained in these folders. We
will also show how to model some of the folders that have to be done by hand.
Important: The interpreters that generate deployment
artifacts expect very strict constraints in the model. When you finish your
model, and any time you wish to generate anything, it is a good idea to check
constraints by clicking on File->Check->Check All in GME. This
will help you find many logic errors in your model.
This folder contains implementation artifacts
associated with components. idl_to_picml has
created these artifacts for us, with their dependency relationships correctly
captured. Figure 4 shows the Implementation Artifacts for Broker.
Figure 4.
This folder contains models that describe the
implementations of component interfaces. In the Quoter
example, we will have two monolithic component implementations - named StockDistributorImplementation and StockBrokerImplementation - and an assembly
component implementation named StockQuoter,
which is an assembly of StockDistributorImplementation
and StockBrokerImplementaion. idl_to_picml has created the monolithic component
implementations for us, as shown in figure 5 for example, but we have to
specify the connections between them in order to construct a Quoter system.
Figure 5
Assembly components provide a boundary for the
composition of monolithic components and even other assemblies. Assembly
components do not provide actual implementations for their interface, it is a
virtual component that delegates its ports and attributes to one or more of the
entities it contains. There are slight differences between modeling an assembly
and modeling a monolithic component. Since the assembly is a virtual component,
it is not necessary to model a MonolithicImplementation
. Instead, instances of other components are
placed within the assembly and connected. Let's try to assemble the StockQuoter.
MyQuoter/InterfaceDefinitions/Broker/Stock/StockBroker
in the browser to the "StockQuoter"
<<ComponentAssembly>> window while
pressing ALT key. Repeat the same steps for MyQuoter/InterfaceDefinitions/Distributor/Stock/StockDistributor
,
and you will see two instances of component interfaces have been created
as shown in Figure 6 (the black line in the figure only indicates
.mapping)
Figure 6
>
Figure 7
This folder contains deployable
component packages. Every instance in an assembly should have a corresponding monolithic
implementation and packaged in a ComponentPackagebe
To create a package for StockBroker.instance
MyQuoter/ComponentImplementations/StockBroker
Implementation/StockBrokerMonolithicImpl.
Refer the <<ComponentRef>> Broker
to MyQuoter/InterfaceDefinitions/Broker/Stock/StockBroker
.
To create a reference in GME, simply drag the tree node in the Browser
"into" the reference model. For example, you should drag the
tree node MyQuoter/ComponentImplementations/StockBroker
Implementation/StockBrokerMonolithicImpl
into <<ComponentImplementationReference>>
Broker. Check GME manual(tutorials) for more information.
The model you built should resemble Figure 8.
>
Figure 8
Now create a ComponentPackage/StockDistributor
following the same steps.
1. Insert a <<ComponentPackage>>
named "StockQuoter"
2. Add a <<ComponentImplementationReference>>
named "StockQuoter", refer it to MyQuoter/ComponentImplementations/StockQuoter/StockQuoter
3. Switch to connect mode and connect <<ComponentImplementationReference>> StockQuoter with <<ComponentPackage>>
StockQuoter, as in Figure 9.
>
Figure 9
This folder contains just one
model capturing specific configuration of Component packages.
MyQuoter/ComponentPackage/StockQuoter/StockQuoter
(Not <<ComponentImplementationReference>>
MyQuoter/ComponentPackage/StockQuoter/StockQuoter
!)
This folder contains one model
capturing information about the top-level element that will be fed to the
application.
Quoter/PackageConfiguration/Default/Default
This folder contains
domain-specific models capturing information about the target environment in
which component-based application will be deployed. In this Quoter
example, we can either deploy the two components into ONE host, or into TWO
different host. We will deploy the Quoter into two
different host.
Now we are ready to deploy our Components to the actual physical environment.
This folder contains a plan model that
captures information about the assignment of component to nodes.
MyQuoter
/Targets/Domain/Broker
and MyQuoter
/Targets/Domain/Distributor
respectivelyMyQuoter/ComponentImplementations/StockQuoterImplementation/StockQuoter/StockBroker
and MyQuoter/ComponentImplementations/StockQuoterImplementation/StockQuoter/StockDistributor
respectively. Note that the <<ComponentRef>>
in DeploymentPlan is not referring to the interface,
but rather the actual instance of the implementation.
Figure 10.
>
Figure 11
Now move your cursor to <<ComponentRef>> StockDistributor,
and left-click it, so that it looks like Figure 12.
>
Figure 12
This operation
associates StockDistributor component to the
<<CollocationGroup>> running on Distributor
node. To do the same with Broker, just right-click on another <<CollocationGroup>>, and left-click on StockBroker <<CollocationGroup>>
which is highlighted.
Up to this point, we are basically
done with the modeling part. Before we generate anything from this model
though, please DO remember to check the constraints of the model as we
mentioned earlier.
After we are done with the modeling, the flattened_deploymentplan interpreter that comes along with
PICML will help us to generate a flattened_deploymentplan.cdp
file. Please make sure you generate this file to MyQuoter/descriptors.
Note that to get the example working,a Node Map file should be created to map logicalnodes to NodeManager
object references,e.g:
DistributorNode corbaloc:iiop:localhost:30000/NodeManager
BrokerNode corbaloc:iiop:localhost:40000/NodeManager
Ming XiongLast
modified: