I have acquired approval from the SourceForge project to have an open source project that is aimed at "computational science education tools and tutorials."  You can access the home page either through SourceForge or directly at http:www.imsa.clemson.edu/OSCOMSETT. Right now the work force has size one :-), but I do have a bunch of stuff collected over the years to support my courses. I intend to put them up over the summer.

I'm a big believer for open source with respect to educational materials but that's certainly not a universal idea. My goal is to have "ready-to-go" materials for lifelong learning.

The following link is a report on evaluation work of the gsoap toolkit for developing web services

Evaluation of gsoap toolkit

We are currently developing a web service application enabling different software packages to communicate data, sometimes these packages may be running on different platforms. This provides a number of benefits for researchers.

• Enables automatic generation of metadata describing different simulations
• Makes a simulation accessible as a web service we can treat the simulation as a service
• Large simulations running as batch jobs can become interactive and users may effectively steer a computational model
• can be used to develop visualisation applications in which multiple users may work collaboratively on a data set

We have been testing the web service client tools provided with matlab. Since our web service applications have been developed using the gsoap toolkit we attempted to use the -m mex option with the gsoap compier and link our client gsoap web service client applications with matlab using the mex. Fearing that this was would be laborious and error prone we reviewed the Web service client tools provided with matlab.

Matlab provides the following Matlab SOAP functions which can be used to develop web service clients

• createSoapMessage
  • Creates the message to be made to the web service
• callSoapService
  • Makes the actual call to the web service using the previously generated soap message
• parseSoapResponse
  • Parses the response from the web service and generates a result

An example web service client call using matlab is illustrated in the following m-file myadd.m. This particular m- file makes a request to the add method of the calculator web service that is provided as an example with the gsoap toolkit examples.

It may still  appear to some developers that it is laborious to develop these client applications but Matlab comes to the rescue with the createClassFromWSDL function. All that is required is the WSDL (Web Service Description Language)  describing the web service that we want to access using the web service client. This generates functions that can be used to call each of the web service methods. This method worked fine for our calculator web service example. 

Matlab creates a folder called @calc (for our calc example) containing the web service methods. When we ran the add example matlab responded with an error 

'Undefined function or method 'add' for input arguments of type struct'

We found that the resolution to this problem was simply to copy the method functions from the @calc directory to the current working directory.

The following slides presents our work on comparisons of linpack

specfp bench mark tests on multi core processor systems. 

An assessment was made of the memory performance.

The results suggest that developers may need to focus effort

on optimisation of memory usage rather than using parallel

development techniques.

Memory Access The Hidden Performance Issues

The iceberg compute cluster is mainly used for high throughput we occasionally find that users applications with a lot of I/O do not perform as well as they should. The application spends most of its time waiting for I/O operations. In such cases we reconfigure the job so that a replica of the data being used by the application is copied to the local hard drive of the compute node used by the application.

The problem arises from the  hierarchy of bottlenecks that arise during the lifetime of an application these bottlenecks can be caused by the latency of read/write operations for

• cache 
• memory
• local file system
• NFS mounted file systems
• Data packets on network fabrics such as gigabit ethernet, myrinet and infiniband.

With the development of iceberg we are facing new challenges in terms of file system I/O it is felt that with greater numbers of users and more compute nodes there will be an increasing strain on the network fabric. Solutions to this problem include using

• dual network cards
• NFS over infiniband
• file replication techniques e.g. exludus
• cluster file systems such as GPFS and lustre

In order to make such decisions benchmarking and monitoring examples include stream for memory bandwidth, IOZONE for file I/O and pallas for MPI benchmarking.

Sustainable memory bandwidth in HPC

ConnectX Infiniband

exludus

IOZONE

HPC using Infiniband

Pallas MPI Benchmark

The Computers & Geosciences journal invites authors to submit papers for a special issue on "Virtual Globes in Science", which will highlight how Keyhole Markup Language (KML) and Virtual Globes have been used for facilitating Geoscience research. This issue will focus on exploring the work highlighted at the Virtual Globes sessions at the Fall Meeting of the American Geophysical Union (AGU), however, it is open also to any innovative contributions on this topic.

Scientific fields of study can include, but are not limited to:

Access grid is more than video conferencing over the internet. This was the lesson learned from a recent visit to the IBM Visual and Spatial Technology Centre at The University of Brimingham (VISTA). Access grid has been used at the University of Sheffield for the last five years. Its main use has been to support the White Rose Grid e-Science centre by allowing its community from the Universities of Leeds, York and Sheffield to meet more regularly.

Access grid sessions can be accessed from a modern desktop for personal sessions in a similar way to Skype and VRVS/EVO. Using a well configured dedicated facility provides a number of benefits.

• Enables participation by a group of people
• Good quality microphones distributed around the venue and echo cancellation enhances the overall sound quality
• A large display using multiple projectors, provides multiple views of the speaker, remote audiences and desktop.

Access grid is a service that is available to to researchers from a range of projects. Researchers may share presentations over Access grid using distributed powerpoint. An additional need for researchers is sharing of applications. For a number of sessions the desktop sharing software RealVNC has been used. This has been used for a presentation on distributed Matlab for sharing visualisation results with IBM data explorer and for sharing Manuscript viewer for high resolution imagery with Flash technology. These sessions have had varying success which is dependent on the degree of interactivity.

Improved shared visualisation has been acheived through the use of customised applications. For example: The Virtual Vellum project has developed a manuscript viewer that allows geographically distributed researchers to interactively control the manuscript viewer. The Magic project has provided shared training material using access grid for a number of mathematics departments in the UK. The University of Leeds have developed visualisation middleware enabling the development of collaborative visualization and computational steering applications for research projects called gVIz.

The use of shared applications is an important feature of a geographically distributed meeting or teaching session. To do this successfully is still quite difficult. On a recent visit to the VISTA centre at The University Birmingham I was given a demonstration of the IBM Deep Computing Visualization (DCV) application. The demonstration involved 3 UK institutions participating in an Access grid session, each one starting  a DCV client accessing the VISTA DCV server. Using a modified version of RealVNC (provided with the IBM DCV package) we were able to share output and control of number of packages using DCV. Interactivity between the three sites was impressive. And the relative ease with which this can be set up seems promising. The following applications were demonstrated Avizo, Google Earth and ArcGIS.

The smart feature about the IBM DCV is its ability to capture the OpenGL instruction stream and to share this with remote sites. In essence the product enables sharing of any applications that use OpenGL.

Our demonstration at the VISTA centre had not ended here we were provided with a demonstration of the immersive sterescopic display donning a specilised pair of 3d glasses we visualised a protein molecule generated by the VMD application. Depth perception was excellent and the capability added to the visualization experience.

The website and abstract submission are open for ISDE6, in Beijing, September 9-12 2009.

http://www.isde6.org/

The 25 November 2008 issue of Eos (which has only just reached me) leads with an article called "Informatics and the 2007-2008 Electronic Geophysical Year" (Baker et al), which concludes with a plea for authors to submit papers that highlight informatics.  See also the eGY website.

(I am collecting a list of relevant journals for earth science informatics at http://delicious.com/jonblower/journals .)

The White Rose Grid iRODS Evaluation and Demonstrator Project This project will provide an in-depth evaluation and demonstration of the iRODS system, assessing its capabilities and role within a distributed data management scheme and complex virtual organisation driven by real-world requirements from the CARMEN e-Science project. Distributed data management issues are still a major challenge in addressing the ubiquity and usability of Grid systems, particularly within the context of complex virtual organisations (VO) where flexible security models and fine -grained role based access are a pre-requisite. The Storage Request Broker (SRB) software from SDSC has established itself as one of the leading Grid middleware applications to support the management of highly distributed large scale datasets for science applications. SRB provides the capability to virtualise distributed datasets, and to provide standardised access to a broad range of underlying storage technologies, spanning flat file systems through to database servers and tape archiving systems. Through the use of SRB, end-users are freed from concerns about the location of data and determining the correct procedures to recall or transfer data to their local or host compute environment. SRB abstracts these challenging aspects of distributed data management away from the end-user, and provides a simplified and uniform way to recall data via indexing systems (metacatalogs) which keep a logical mapping of the underlying distributed data. SRB has been widely adopted within large-scale Grid applications, particularly in the science communities, and provides the data management backbone for the National Grid Service (NGS). However, SRB does have limitations particularly in regard to integration with complex VO’s. These have been recognised by the developers of SRB at SDSC and a new version, iRODS, has recently been released to address the short-comings within SRB. Many of the limitations within SRB relate to the metadata schemes used to describe and annotate the data collections. In many cases these metadata mechanisms are too restrictive and inflexible to support complex meta-data schemes for complex role-based Grid systems. iRODS is dowloadable at release 1.1, and as part of the objectives of the JISC e-infrastructure imitative, this project will assess the impact that the new features and functions within iRODS will have on the UK e-Science community, and assess its potential for deployment within the NGS. [Note that the current iREAD demonstration uses an iRODS pre-release of 1.2 kindly provided by UCSD.]