[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [sumo-user] statistics output & summary output
|
Thanks for clarifying things, Jakob.
First things first: I considered the settings in sumocfg-file as added to those that were supplied via command line arguments. Indeed, this seems to apply partially. I have <duration-log-statistics value="true"/> in the sumocfg file. When I omit this, then a short stats.xml is generated (upon command-line invocation with the -c argument).
When I do "reload" in the GUI, then the command line parameters are still active. And according to my understanding, reload starts parsing the sumocfg again.
Regarding handling/workflow:
I usually start sumo-gui from the desktop, and then load the config file via the file menu - just to 'mimic' the typical windows user would do it the same way. I only use the shell when I - sometimes - want to activate some other behaviour (e.g. bypass default teleporting).
Because of the precise behaviour you outlined now, I have decided to tell the linux system that upon clicking on a sumocfg-file, this file shall be opened with sumo-gui. Even more windows-like & appropriate, but also more convenient if I simply click on a sumocfg-file that implies other behaviour or one that is decicated to graphic edgedata evaluation.
For the sake of completeness: moving the --statistic-output <file> to the sumocfg indeed enables statistics output when invoked from the desktop. But it results in a stats.xml with absolute paths in the <input> section (all additional files etc.). Also when I click the sumocfg in the file manager (in the working directory, and all paths in sumocfg are relative - version 1.12.0+1429).
BTW: I'd suggest to output stats content also as a simple small table after simulation end - e.g. as an extension of the dialog box that pops up at the end of the simulation.
Regards,
Rob
Am 17.03.2022, 13:57 Uhr, schrieb Jakob Erdmann <namdre.sumo@xxxxxxxxx>:
I did not realize you were opening config files in sumo-gui.
This is the expected sequence of events when you load a new .sumocfg from within sumo-gui:
- the simulation that was started from the command line is aborted and any opened files are flushed and closed
- a new simulation is loaded that only uses the options defined in the loaded .sumocfg (prior command line options are not applied!)
All options that work from the command line also work in a .sumocfg so you could
include the line <statistic-output value="output/stats.xml"> to get your output when loading a config in the gui.
regards,
Jakob
(please forward a few lines below for my remarks)
> Therefore I point out mor clearly that without using the -c command line
option, no summary is generated. At the moment I cannot say whether
this condition also applies to statistics output, since I did the tests
on my office computer.
I'm pretty sure that this is working in our test. Please post the exact command that doesn't behave as expected.
$ sumo-gui --statistic-output "output/stats.xml" --summary output/summary.xml &
With this command, only edgedata.xml is generated (right from the start, just after having loaded the sumocfg file from the file menu. Note that edgedata.xml is specified via an include of an additional-misc.xml file with following contents:
<additional>
<edgeData id="full" file="output/edgedata.xml"/>
</additional>
Of course, edgedata.xml is also generated when sumo-gui is not started from the command line.
The report-section of the sumocfg file contains this:
<verbose value="true"/>
<duration-log.statistics value="true"/>
<no-step-log value="true"/>
Therefore I'd expect that a statistics output is generated. But I think that a missing statistic-output argument makes that this is not the case. Therefore I was using the commandline interface, hence the --statistic-output in the quoted command line above.
When I invoke this...
$ sumo-gui -c osm.sumocfg --statistic-output "output/stats.xml" --summary output/summary.xml &
... both stats.xml as well as summary.xml are created right after the start (initially only containing the header).
Hence, indeed the statistics output is generated only when specifying the osmcfg file via the command line.
Tested with 1.12.0 and also with 1.12.0+1158.
At any case: only when a file is present with header content after loading the simulation, the file will grow (summary) during or show content (edgedata, stats) after a simulation ends: there will by no output file when it it not present from the start. This allows very quick determination of behaviour without actually starting the simulation.
Regards,
Rob
> Regarding buildup of summary data in memory
Summary output size grows linear with simulation duration and would therefore function as a "memory leak" in long-running simulations if not flushed to disk. In contrast, statistic-output uses a fixed amount of memory since it only accumulates in a constant number of variables.
Jakob,
perhaps I was a bit misunderstanding. Therefore I point out mor clearly that without using the -c command line option, no summary is generated. At the moment I cannot say whether this condition also applies to statistics output, since I did the tests on my office computer.
Regarding buildup of summary data in memory (array) or disk: Well, let me compare with statistics output. This is something that IS build up into memory and is flushed into the file at the very moment of closing the final simulation end dialog with "yes" (closing view & files). Whether it's only flushing the file output buffer or outputting a data structure at that time is secondary.
Regards,
Rob
The documentation already says "The elements vehicleTripStatistics, pedestrianStatistics, rideStatistics and transportStatistics are only generated when either of the options
--duration-log.statistics or --tripinfo-output are set."
I've now put this into a yellow box to make it more visible.
Note: the reason why this isn't on by default is because the additional bookkeeping required by this feature can slow down the simulation.
> I'd suggest to accumulate the summary in memory and output it at once to file, because this wouldn't slow down simulation.
I'm not sure I understood the suggestion. Do you propose that accumulating in memory would bring a speedup or that it would bring some other benefit (without slowing down the simulation)?
best regards,
Jakob
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user
--
Dem Ingeniör ist nichts zu schwör (außer bei Force Majör)
