pw, I remember that you in one of the messages marked what to switch between technologies rather simply. if it concerns a line similar - "scrypt": true, yes, but is features with thin setup of frequencies, thread-concurrency, etc. didn't reflect?
Not between technologies (algorithms I assume you mean). Only between Scrypt-N nfactor values. There might be support for switching algorithms built into a miner later, but I'm not sure they exist yet (I haven't dug that deeply though). And if parameters can't be specified per-algorithm, you're going to be mining pretty sub-optimally a lot of the time.
Indeed. There's no publicly available miner yet that support multiple algorithms and hotswapping. Implementing X11 and a few other algorithms on top of the latest sgminer is reasonably trivial (the X11 patch from sph-sgminer is fairly clean, and isn't difficult to modify to use the sgminer algorithm pattern vs only the classic kernel pattern). However, I have seen no work done on making parameters algorithm specific.
If nobody else tackles it by the end of the month, I plan on cleaning up and releasing my own sgminer derivative (maybe clgminer? or the ever vain phziner hah) with multi-algorithm hotswapping support. Algorithm-specific parameters and stratum extensions are my hold-back at the moment.
@PoolWaffle: maybe if you're interested, we could plan to launch a multi-algorithm GPU-only version of WafflePool simultaneously with clgminer. The miner would be open source, but not the server-side stratum layer, so it would give WafflePool a solid first-to-market opportunity.
I have been fighting with myself a bit on how to handle the fact that different GPUs perform differently on various algorithms - e.g. one GPU might do 500KH/s scrypt, 250KH/s scrypt(n=11), and 1.5MH/s X11; whereas another might do 400KH/s scrypt, 300KH/s scrypt(n=11), and 2MH/s X11. So you have a 4:2:6 ratio vs 4:3:10 ratio. This is fairly significant, as at a given moment, the first rig might be best hashing scrypt(n=11), and the second would be most profitable to hash X11. So, some sort of local benchmarking with relative performance report back to the pool is necessary to optimize profits - of course, this would complicate switcher logic significantly... Power consumption is also different between algorithm (which some users will care about, and some won't), so a user-specified algorithm weight preference might be the easiest solution.