load average graph scaling is misleading

Bug #802453 reported by Steven Noonan
14
This bug affects 3 people
Affects Status Importance Assigned to Milestone
System Load Indicator
Fix Released
Medium
Michael Hofmann

Bug Description

The minimum value currently in place for indicator-multiload's load average indicator is '1', which makes a load of '0.25' (or other such small load) appear far more severe than it actually is.

I (and probably others using this indicator) am used to the way gnome-applets scales it. They actually have a hard maximum of '5' set, which seems to work rather well: http://git.gnome.org/browse/gnome-applets/tree/multiload/linux-proc.c

I suspect the "correct" solution would be to set the minimum to the number of logical CPUs or something. On a 12-core hyperthreaded machine (i.e. 24 thread), a load average of '0.25' is *nothing*, and the indicator should reflect that. However, on a single-core, single-thread machine, a load of 0.25 might be indicative of something impacting user-noticeable performance.

Revision history for this message
Michael Hofmann (mh21) wrote :

Sounds good. A minimum corresponding to the number of cores, but what about the maximum? Would it still make sense to automatically scale the graph per default (there will be an option to configure the scaling to your liking in dconf eventually)?

Changed in indicator-multiload:
status: New → Confirmed
milestone: none → 0.3
importance: Undecided → Medium
Revision history for this message
Steven Noonan (steven-valvesoftware) wrote :

(Sorry for delayed response, didn't get email notification for some reason.)

The maximum could scale, but unless it's somehow made obvious that it's scaling, it'd be kind of pointless. The main issue is that the graph needs to convey useful at-a-glance information. The dropdown should be available, but not -necessary- to interpret the data. If the graph looks like it's peaking, that should be an indicator of high utilization of that resource. (I don't recall what I saw in the code, but I think the network graph scales the same way. On an idle internet connection, it would occasionally peak because some Windows machine sent a broadcast for SMB.)

I suspect that for system load, it might be reasonable to just not scale it at all. Gnome multiload-applet doesn't scale it, I don't think.

Michael Hofmann (mh21)
Changed in indicator-multiload:
assignee: nobody → Michael Hofmann (mh21)
status: Confirmed → Fix Committed
Revision history for this message
Michael Hofmann (mh21) wrote :

Should be available in the beta versions at ppa:indicator-multiload/daily in a couple of hours (0.3-0~37~15).

With the new version, you can specify the minimum and maximum for the upper graph edge with dconf-editor.

The default for the minimum of the load graph is the number of cpus.

The maximum is still set to scale automatically. If it is annoying though, don't hesitate to open another bug report :-).

Revision history for this message
Steven Noonan (steven-valvesoftware) wrote :

I think the biggest issue was the maximum scaling down, which made the rather low load appear much more catastrophic. :)

Michael Hofmann (mh21)
Changed in indicator-multiload:
status: Fix Committed → Fix Released
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.