hashview-old
hashview-old copied to clipboard
hashview
Relative chunk distribution for agents
First off, awesome meeting you guys at BlackHat!
So, we currently have several machines running agents (~250 GH/s total if I recall correctly) checking into the central Hashview instance. We notice that when an agent device joins, a benchmark of NTLM is run and then added to the running combined total -- each of varying individual power. When you start a new job, the workload of each task is distributed equally based on the variable chunk number defined in the global settings. The problem here is, for example, that if I have 2 devices (central and agent) that total 100 GH/s, but individually they execute at 90 GH/s and 10 GH/s, they both appear to take responsibility for equal chunks. this leads to a problem towards the end of the task, as the slower machine may still be working on a chunk long after the faster machine has exhausted the remainder queue. What would be awesome would be if the chunking was 'smart' and could identify that one machine is responsible for 90% of the total combined efficiency and the other was only 10% of the efficiency that it would distribute the chunks in such a way where theoretically the loads were relatively equal. What this would allow over the longer term would be the ability to deploy agents on every machine in our arsenal, even laptops or workstations with low-mid end graphics cards, knowing that while they won't be a huge contributor can still pull their relative weight over weekends or holidays when we decide to 'check them in'. I've said it before, but think SETI, but for hash cracking. ;-)
As always, I greatly appreciate the hard work you all do, the openness to community ideas/input, and look forward to each new feature you roll out!
