Joined: 6 Oct 06
Am I correct in thinking this is still in effect?:
"On the other hand, if there's a single cross-resource LTD, and only one project has GPU jobs, then its LTD would go unboundedly negative, and the others would go unboundedly positive. This is undesirable. It could be fixed by limiting the LTD to a finite range, but this would lose information.
In the new model:
There is a separate LTD for each resource type
The "overall LTD", used in the work-fetch decision, is the sum of the resource LTDs, weighted by the speed of the resource (FLOPs per instance-second)."
Just wondering, what is undesirable about just having one single debt for the whole host? Say I have Einstein (which does GPU and CPU) and Rosetta (which only does CPU) attached to a host, with equal weighting. What's wrong with it running Einstein on the GPU all the time (the only thing possible), then causing Rosetta to have a huge debt, so it always gets the CPU? That is afterall as close to 50% of each as I can get on that host. It seems like it will use half the CPU resource for Einstein, which it shouldn't, since it's getting all that GPU work done.
Copyright © 2021 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.