I'm trying to simulate settings with 10 casus, and 30-54 bees. I am using an 8-core i7 3.6GHz, which should be sufficient for such simulations. However, the simulator reports 10-30% slowdown in the 30-bee experiments and higher with more bees. This is experienced throughout a simulation; in prior experiments with up to 36 bees and 6 casus I would typically only observe a high skew in the first measurement period, when the objects are being spawned, and afterwards the load would reduce.
I'm looking for proposals that can reduce the load, or at the very least diagnose the load to identify the problems.
One item we discussed previously was to implement a 2nd type of bee that has fewer IR sensors on it - if the implementation does not use all the sensor information then we could save a % of effort this way. (It is clearly only a linear patch, not fundamental improvement in scaling; but right now I'm looking for pragmatic improvements)
I'm trying to simulate settings with 10 casus, and 30-54 bees. I am using an 8-core i7 3.6GHz, which should be sufficient for such simulations. However, the simulator reports 10-30% slowdown in the 30-bee experiments and higher with more bees. This is experienced throughout a simulation; in prior experiments with up to 36 bees and 6 casus I would typically only observe a high skew in the first measurement period, when the objects are being spawned, and afterwards the load would reduce.
I'm looking for proposals that can reduce the load, or at the very least diagnose the load to identify the problems. One item we discussed previously was to implement a 2nd type of bee that has fewer IR sensors on it - if the implementation does not use all the sensor information then we could save a % of effort this way. (It is clearly only a linear patch, not fundamental improvement in scaling; but right now I'm looking for pragmatic improvements)