Technical Overview of WebCom
| The Gander audit management solution transparently harnesses WebCom cutting edge distributed computing technology. WebCom is a distributed computing platform which is the product of over 10 years of research and countless man hours of development at the Centre for Unified Computing in University College, Cork It is a workflow execution engine which can transparently harness various resources at runtime, according to various policies. WebCom is composed of a number of modules which manage the different aspects of WebCom's functionality. |
 |
Compute Engines: Execution of tasks within an application workflow is performed by a compute engine.
All WebCom instances typically run with a Condensed Graphs engine module which allows the execution of
applications or workflows described as Condensed Graphs (CGs). CGs allow the developer to implicitly or explicitly
utilise task constraints and dependencies, parallelism, lazy or eager computations within the same computation, n
levels of abstraction, and any number of underlying technologies. To extend WebCom's computational abilities,
additional compute engines may be loaded for user defined instruction types or to take advantage of different
technologies or hardware. Examples of additional engine modules include an ORTE Engine Module to allow WebCom to
target work to the ORTE MPI management system for HPC tasks and a Minimum Intrusion Grid engine module to allow
WebCom workflows to harness MIG resources.
When an audit is to be performed by Gander, an application workflow for the audit is automatically generated
according to the users' infrastructure and audit preferences. The WebCom engine executes this workflow by
automatically exposing sub-tasks (in parallel where available) and either scheduling them to appropriate resources (e.g.,
machine, application, admin, CEO) or executing them locally.
Connection Manager: WebCom can load any number of connection managers which handle connections and the transfer of messages between WebCom instances. A single WebCom instance can utilise a number of connection managers simultaneously, to allow the integration of different protocols and technologies into the enterprise audit project.
Task Distribution: The distributor module is responsible for allocating and scheduling work between collaborating WebCom instances. Its decisions are guided by policies associated with applications themselves, and are bounded by site policies put in place by resource owners. Policies can specify scheduling constraints and actual algorithms for choosing amongst suitable destinations for work packets. Gander utilises this functionality to target audit subtasks between collaborating Gander systems and users.
Security: Security is built into WebCom at various levels. SSL and certificate based authorisation for users and resources are harnessed by a Keynote trust management system which allows the application developer to specify security constraints in the application workflow. Security policies are also specified by site admins to constrain who or what may target their resources.
Fault Tolerance: The Fault Tolerance Module deals with faults in a decentralised manner. Should an active WebCom resource fail, its local neighbours collaborate to reconfigure the network topology and rebuild the computation in that area. A roll-forward mechanism is used to quickly restore the execution state to that which existed prior to the failure. Fault tolerance housekeeping may be automatically throttled in order to deal with increasing expectations of faults or according to the reliability reputation of resources in the WebCom world.
These core modules can also be supplemented by User Modules, which implement any functionality outside of the core domain.