Interpreting the timeline in Platform analyzer?

Hi,

I am running my OpenCL application on an Intel HD 530 graphics device and experience exessive time usage for some kernels. On all previous devices I have tried (including earlier Intel HD), the measured time of my kernel during repeating calls has been fairly constant. On the HD 530 graphics, the execution time jumps between 3ms (normal) to 40ms(!). 

I got two questions:

1. In the attached sceren shot, the Platform Analyzer shows me the execution times for my kernels, but I don't know how I should interpret this, as each kernel is displayed several times, on different lines, with different execution times. Is the top line of colored boxes, the "actual execution time" and the lower ones indicate when the command was issued from CPU?
2. Do you have any idea why my execution time would jump between 3ms, which is the normal, and 40ms? When I enable more computation in my kernel I can see the 3ms is increasing, but the 40ms is constant. As I mentioned above, this has never happened on other devices (NVIDIA, AMD, Intel), which makes me think there is something else going on on the GPU.

Thanks