New features
------------

Some thoughts for new features which I hope to get around to.
See also BUGS which probably need fixing first.

1. Often we want to look at behaviour over time when we change
something, so cumulative statistics are not ideal: we'd like to see
a window of data which gradually moves over time. Look at how to
implement this with ps_top.  I'd expect the window to be rather
small, perhaps 1 minute, or maybe longer triggered by collecting
data less frequently.

2. vmstat, iostat, pt-diskstats and other such utilities show values
over time but providing output as stdout. It might be useful to
have options to pstop to present the info in that sort of format
too.

3. Add some replication metrics to pstop as most of my servers are
slaves and we want to know where the server is busy. I've seen a
way to do that and need to incorporate it into pstop.

4. _if_ ps statistics get truncated it would be good to be able to
remember the latest state and add those metrics back again, _if_
looking at absolute rather than relative statistics. On servers I
manage I currently truncate info as it is collected after 15-minutes,
so this might happen while watching a server with pstop.

5. Improve the options for sorting the data. Total latency is good,
but it would be good to see the write based metrics ordered by total
write latency. So look to improve the display and sort options here.

6. I/O Latency by File shows read/write bytes and operations but
none of these values is converted into B/s or ops/sec which might
be better given we want to see the rate values.  MB/s or ops/s are
easier to understand and the values don't change depending on the
period being viewed. That said unless we have the times the metrics
were collected we can't calculate these rates so the ABS values
can't show the rate information but would need to show only the
counters.