GPU + CUDA possibilities

I thought (am am willing to be proved wrong of course) that the GPU excels at parallel tasks and moving blocks of memory. Things like a brute force attack probably fit into this area nicely, but it's hard to know what other sorts of tasks require a lot of processing in this manner.
Disk I/O is a big part of our world, and your GPU isn't helping you there.

If you haven't already, read this http//www.cs.uno.edu/~golden/Stuff/dfrws2007-gpu.pdf

Personally, I think that there is a lot to be gained from investigation of parallell computation techniques in Forensics - otherwise we are going to be hitting I/O bottlenecks when attempting to do analysis of even the smallest systems - I'm not sure that GPGPU is going to help - as Gromit points out, this doesn't help in the I/O issue - but with more traditional clusters or parallell machines, I think there is a _lot_ of room for research …

I know its a little early, so pardon the stupid question, but what type of situation do you find disk i/o to be your bottleneck? You might want to look into some faster drives, and/or more spindles. Just shooting from the hip here as I haven't done any tests or know what kinda software you are running.

Sorry, I'm speaking in terms of theory rather than practice …

However, I'd suggest that already disk is a bottleneck as it is by and far the slowest part of the system ( screen excluded ). Just we don't notice, because the overheads of everything else are so daft ! If for example, I ditch the ridiculous requirements of graphical user interfaces ( and by that I mean both Windows and FTK/EnCase ) and do a simple grep on a disk image for a string using an optimised Linux kernel, I expect that my multicore cpu would be nowhere near 100% load, where as my disk I/O would be maxed out. The time that this would be most noticeable is when you are indexing your disk initially - and as we all know this isn't a particularly quick process !

Taking the fastest processor one could, theoretically, pass data at speeds in the order of tens of gigabits per second (Intel is claiming 32GB), however, even with the fastest disk I/O ( solid state disks ) coming in at a shade over one gigabit per second there is obviously a huge disparity. Caching helps to reduce this difference when we are doing things at human speed in actually performing an examination, but still if you decide to do a keyword search it is considerably slowed by disk I/O.

I must admit that I've just Googled these figures, I'd made my statement that disk is a limiting factor from years of experience - not in forensics, but in running high-end multiprocessor UNIX servers and clusters for databases and scientific applications - showing that even where the processor is maxed out doing mathematical computations 100% of the time, disk/memory/network I/O is still the limiting factor.

I hope that I've explained myself adequately !

Hey sleepy! i dont know if you remember me but i was from the original thread. Your original suggestion to do a project on distributed grid computing to break certain encryption standards. Well the original project that i decided to do is finished and over with. Its now time for another, well it soon will be D . Anyway this project will be my final year project. The project only has to be in next year, but theres nothing wrong with generating ideas. And i found your grid computing idea very intresting.

Iv had some experience breaking passwords and stuff like that, i would also say iv had some C++ programming experience (years & half), done some research into grid computing and MPI.

This seems to be the suject topic to post to. I would be making the project from scratch or almost using C++. So i was wondering what to look into further; programming wise, weather it is feasable, well genrally starting points and futher down the line a bit more advanced stuff.

Thanks

thegavster