Possible libpcp interpolation issue

[myllynen]

Using https://fedorapeople.org/~nathans/archives/goody.tgz we see with the latest git code:

pmrep -a 20160718.12.38.0 -z -S '@19:45' -s 20 -J3 -2 -u proc.io.read_bytes 0.05s user 0.01s system 99% cpu 0.069 total

pmrep -a 20160718.12.38.0 -z -S '@19:45' -s 20 -J3 -2 -t 1sec proc.io.read_bytes 0.19s user 0.05s system 99% cpu 0.252 total

pmrep -a 20160718.12.38.0 -z -S '@19:45' -s 20 -J3 -2 -t 10min 2.42s user 0.53s system 99% cpu 2.967 total

Adding a simple print debug in the Python code (around line 1145 in the beginning of the fetch() method) we see that the number of C calls for each is the same (as expected due to -s 20). The first one disables interpolation (-u) but the next two use it with different interval (1s vs 10min).

The question is why the -t 10min case is so much slower than the -t 1sec case, is there anything that could be done to improve this?

Thanks.