Summary of related works of thesis proposal:

Parameter Sweep

Paper (Cached copy)SummaryRelation to thesis
NimrodTool for Parametised Simulations on distributed systemsSimilarities: Parametised Simulations on distributed computers. Differences: No active sampling. Designed for large, widely distributed computational systems, not clusters. Designed for fault tolerance, not interactivity.
Nimrod/OSimilar to Nimron, but with automatic optimisation.Similarities: Guided search of design space. Differences: Not focused on interactivity - no consideration to partial results. The objective can only be expressed by a single objective functions.
CondorA cooperating distributed batch system, deals with job planning scheduling and resource management.Focuses on flexibility - emphasis is placed on dealing with volatility of computing resources.
GlobusToolkit deployed on grid resources for automatic resource discovery and workflow scheduling.It can be used for distributed parameter sweep applications, has a notion of dependence, but is targeted to widely distributed networks, with focus on discovery, data replication and fault tolerance, rather than interactivity.
NetSolveClient-Server system to distribute computations over a network.Addresses the issues of load balancing and resource management, but resource as fixed - i.e., each machine has a fixed number of processors, so the scheduler cannot choose to schedule a portion of a machine.
VI-SteeringInteractively steerable parameter sweep applicationSimilarties: user directed parameter space search on distributed computing resources. But interactivity is limited to the user specifying spaces of interest, rather than steering the entire study. I.e., VI uses a user as the sampler.

Parallel runtime

Paper (Cached copy)SummaryRelation to thesis
Fine-grain dataflow thread schedulingA simple model to describe fine-grain scheduling on multi-threaded multiprocessorsThe scheduling of "quanta" provides way to model and schedule dataflow-based computation on distributed resources. But it does static analysis, i.e., all the pieces of the program has to be known before hand.
ICC++Heirarchical approach to load balancing threads on a distributed platform. Groups threads into subsets and load balance each independently.Heirarchical approach simplifies the scheduling problem on a distributed platform, but requires user intervention to identify subsets and give scheduling policy for each subset.
Affinity SchedulingAllow programmers to specify functions' affinity to tasks, objects, processors, and also allow programmers to specify how objects are distributed.One way to express thread locality (via data and processor locality), but requires programmer intervension.


Paper (Cached copy)SummaryRelation to thesis
FalconOn-line monitoring and steering of large-scale parallel programs.Devised a mechanism to monitor and steer running programs. But the system steers only one experiment at a time, rather than entire computational study. (Did not use because Falcon requires applications to be written in the CThreads package.)
Mirror Object ModelModel for monitoring and steering parallel programs. Object-centric view: Steering and monitoring are performed on objects states and methods placed in mirror objects (or stubs). Provides one mechanism of steering and monitoring, but steers only one object at a time, rather than entire computational study.
Distributed LaboratoriesPresents a multi-user steering platform. Extends the Falcon work by adding two steering/monitoring mechanisms (synchronous and asynchronous); addresses issues of having multiple users - synchronisation of displays, interconnectivity, and data exchange.
SCIRunInteractive computational steering environment using dataflow model. Component-based infrastructure, but focused on the dataflow model and interactivity, and not targeted on distributed platform.
SCIRunSame as aboveSame as above
SCIRun DistributedEnable SCIRun components to be run on distributed machines via stub.Still focused on dataflow model, and not sure how scalable this is.
CUMULVUSInfrastructure for steering, monitoring, and checkpointing.Data-centric view: only parameters are steerable. Monitored and steered variables are declared in advance (using instrumentation code), along with data layout (for distributed objects). Merges visualisation and steering with checkpointing, since they require similar infrastructures.
CUMULVUSCUMULVUS with collaborative analysis.Same as above
UNITAHComponent-based Problem Solving Environment on distributed platformsUse CCA components on the SCIRun dataflow model. Interoperability via adapter components and Nexus communication layer. Supports distributed components. Requires external steering-aware runtime environment.
DISCOVERWeb-based collaborative problem solving environment, allows online steering and monitoring of experiments through the web. Leverage Web-based technology to construct control and monitoring network.
CSESteering and monitoring of computational processes on remote computers. Data-centric in that all operations between the simulation and the user are based on data. But the data are functionally grouped together using Satellites.
RealityGridSteering and monitoring on remote computersSteering library is seperate. Also Data centric - all changes are done via changes in parameter values. Added notion of using checkpoints to support rewinding. Also depends on external runtime environment, in this case Open Grid Services Infrastructure.
RealityGridSame as aboveSame as above.
WEDSLightweight wed service-based middleware focused on transparency: All-in-one middleware to handle job submission, steering, visualisation and workflow manipulation. Based on WSRF services.
GVIZVisualisation middlewareFocuses on visualisation. Also web-based, parameters are sent via XML and SOAP (eventually, using Unix socket calls as of 2003).
SCIRUN2Workflow-based CCA Problem Solving Environment. Ideally, pieces of SimX will be transformed into components in SCIRun2 to enable reuse and quick construction of steerable studies.
CCAStandard interface for Components used in High-Performance Scientific ComputingSame as above
CBPSSummary of SCIRun, BioPSE, and UnitahSee SCIRun, BioPSE, and Unitah