edikt2 Tech Meeting Notes
4 September, 2007
Medical Education Centre, Western General Hospital
AGENDA
1330 Welcome
1335 Data Management in Brain Imaging
- Dr. Trevor Carpenter
1405 INWA - Customer Data Analysis over the World-Wide Grid
- Mr. Terry Sloan
1435 COFFEE
1500 FPGA
- Dr. Rob Baxter
1530 ECDF update
- IS Team
1630 Discussions and Wrap-up
1700 CLOSE
Data Management in Brain Imaging
Trevor gave an overview of the issues faced by SBIRC (SFC Brain Imaging Research Centre) in terms of how they collect, store and manage the vast quanitities of data produced. Issues of anonymity, storage of meta-data and archiving were raised. A new project, SINAPSE will use the DICOM system for storage. The Grid will probably play a key part in this project.
Ewan Roche asked about reclassification of conditions (which are described in the meta-data). This is not really possible.
Rob Baxter wondered what level of anonymisation is required in the SINAPSE project. The plan is to use end-to-end encryption with some anonymisation possible. There are issues with DICOM which can only store 64 characters for a name, but 128 or 256 characters would be better for strong encryption. The hope is that different levels of encryption can be used which gives access to different levels of data. For example, having key with level 0 will give full access whereas a key with level 1 will not give access to the name.
Checksums were also discussed. A simple chacksum for the whole file cannot be used, as even if nothing is changed, simply copying over the file would change dates within the file, hence the checksum would be incorrect. A cleverer checksum is required.
INWA
Terry's talk was a version of a talk given at the recent APAN conference. It dealt with data mining using grid technologies, in particular OGSA-DAI. Interesting path of project has gone from both analysts and data in Edinburgh to tri-continental scale (UK, Australia and China).
Andrew Turner asked about time limits on certificates. Proxy can have a time of up to a month. Longer time is less secure. If certificate expires before a job completes, data should not be lost.
FPGA
Rob gave a brief overview of FPGAs and their potential use. The project has ported three scientific programs to the FPGA cluster. All showed increased performance on the FPGA. Effort required to port is a lot.
Andrew Turner asked why FPGAs are used in devices, rather than convential chips. Reason is that they can be updated with new programming on the fly.
Mike Baker asked about the reliability of benchmarks given that an FPGA should be more predictable than a conventional processor. In principle, yes, but devices are getting bigger and hence more "error" has crept in.
Andrew Turner also asked about the role of FPGAs. Co-processor or main processor? Both AMD and Intel seem to be looking into using FPGAs as a co-processor with fast connection to convential processor. This is not currently feasible due to having to go over PCI bus to get to FPGA.
ECDF
Ewan gave an update on the status of the ECDF and some plans for the future.
Andrew Turner asked about current libraries that are installed. Most standard libraries are available, including FFTW.
A heavy user of Eddie then gave a talk. Partha Lal is a PhD student in informatics looking at advanced speech recognition.
Ewan Roche asked how long each utterence takes to decode. About an hour.
Andrew Turner asked if other languages (than English) are harder to recognise. Apparently not, but the model used may need to be tweaked.
Other Items
Ewan Roche asked about an edikt2 PowerPoint template. Jon Hill will upload it to the wiki soon.