Does anyone have insight into how the higher data rates of the Mk2 devices will impact compute requirements (or even the compute requirements of Mk1)? One of the few papers that deals with this question (that I know of) is here
http://bioinformatics.oxfordjournals...rmatics.btv046
and here
Unless I am mus-interpreting something in the paper, HMM is pretty computationally intensive and seems to require something on the order of 1 mid-range core for every stream of 100 bps or so. Even if this estimate is inefficient by a factor of 1000 the compute requirements of a Mk2 Minion or a Promethion look fiendish. If the devices are trully capable of "run until" as ONT claims them to be
Minion = 3000 pores x 500 bps = 15 cores assuming that the above estimate is 1000x too slow
Promethion = 144,000 pores x 500 bps = 720 cores, again, assuming the above estimate is 1000x too slow
These are impossibly high numbers, so something seems off. Can any current MAP-er comment on their understanding of the compute overhead? I understand things are in the cloud so it may be difficult to guess, but roughly how long does the base calling analysis take and, if you have an idea, how many instances does it run on AWS?
http://bioinformatics.oxfordjournals...rmatics.btv046
and here
Unless I am mus-interpreting something in the paper, HMM is pretty computationally intensive and seems to require something on the order of 1 mid-range core for every stream of 100 bps or so. Even if this estimate is inefficient by a factor of 1000 the compute requirements of a Mk2 Minion or a Promethion look fiendish. If the devices are trully capable of "run until" as ONT claims them to be
Minion = 3000 pores x 500 bps = 15 cores assuming that the above estimate is 1000x too slow
Promethion = 144,000 pores x 500 bps = 720 cores, again, assuming the above estimate is 1000x too slow
These are impossibly high numbers, so something seems off. Can any current MAP-er comment on their understanding of the compute overhead? I understand things are in the cloud so it may be difficult to guess, but roughly how long does the base calling analysis take and, if you have an idea, how many instances does it run on AWS?