ModelInference / synoptic

Inferring models of systems from observations of their behavior
Other
81 stars 25 forks source link

Investigate adding support for non-monotonically increasing series #361

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago
Perfume currently requires event performance metrics not only to be totally 
ordered within a trace but also to be monotonically increasing. In other words, 
the metric cannot decrease from one event to another; it can only increase.

This trace is fine:
A 0
B 5
C 10

While this trace is not:
A 0
B 5
C 1

More accurately, the second trace is accepted but reordered so that the order 
is A0,C1,B5. This doesn't allow Perfume to support some metrics. An example is 
CPU utilization, which will naturally fluctuate up and down (1% at idle, then 
99% during heavy computation, then back down to 1%). Other metrics of this type 
include memory usage, network transfer speed, and disk read/write speed.

Further, negative metrics are not allowed at all right now. E.g., the first 
line in this trace is rejected by Perfume:
A -5
B 0
C 5

This task will involve investigating to what extent the Perfume code depends on 
positive, monotonically increasing metric series. What parts of the code need 
to be changed to allow negative metrics and decreasing series? Is there any 
refactoring necessary to make these possible?

Original issue reported on code.google.com by tonyohm...@gmail.com on 3 Feb 2014 at 7:58

GoogleCodeExporter commented 9 years ago
To clarify, we ultimately want to support a trace like this:

(event, memory footprint)
allocate 5
process 100
send 100
cleanup 110
exit 5

Or this:

(event, net bytes sent)
start 0
login 0
send 15
receive 115
logout -85
exit -60

It's important to note that the code currently orders events in a trace by 
basically sorting the metrics, ignoring the order they show up in the log. To 
allow decreases in performance metrics, we'll need to change this so that trace 
events are ordered by log position.

Original comment by tonyohm...@gmail.com on 4 Feb 2014 at 12:22

GoogleCodeExporter commented 9 years ago

Original comment by tonyohm...@gmail.com on 24 Feb 2014 at 1:18