4 ms·
I have a similar story but I was not quite so clever. Back in 1990 and 1991, I used a cluster of IBM RS6000 workstations in my university to train neural networ
by 2sk21 5y ago
I have a similar story but I was not quite so clever. Back in 1990 and 1991, I used a cluster of IBM RS6000 workstations in my university to train neural networks. I had previously tried to get backpropagation to work on a Connection Machine but found it too frustrating to work with. The cluster of RS6000s actually ran my code (written in C) very fast and was able top get some good results and graduate with PhD. I built a distributed queuing system to ensure that each work station would pull the next job from a central queue.
- exikyut 5y agoDo you remember why the CM was so frustrating? That's a very ironic story right there.
- 2sk21 5y agoMain problem was that it was hard to get time on our CM2 to run jobs. You had to sign up in advance and it was a nuisance. Also debugging the parallel version of C for the CM2, called C* was pretty hard to debug as well.
- eps 5y agoThere used to be something called PVM - Parallel Virtual Machine - a fairly simple library for distributed computing. Had to use it once back in Uni, it wasn't bad. Edit - http://www.csm.ornl.gov/pvm/ http://www.csm.ornl.gov/pvm/
- 2sk21 5y agoYes PVM would have been great had it been available but as I recall, it came out later, in the mid 90s.