Question about trend graph scaling
Hi all,
When I go to the trend page for a host and then click on a graph to get the 48-hr, 12-day, 48-day, and 576-day graphs, something doesn't look right.
For example, I'm looking at the CPU utilization graphs:
- In the 48-hour graph, I see 2 spikes for "user" this morning that were near 60%.
- In the 12-day graph, those same 2 spikes only go up to about 30%.
- In the 48-day graph, the spikes only go up to about 15%.
- In the 576-day graph, the values hover around 10%.
Furthermore, the legend at the bottom of the graphs shows max values that are lower than what can be seen on the graphs. I'm seeing the same behavior on the other graphs from the trend page...not just the CPU Utilization graphs.
Is there a problem with the scaling of the y-axis between the graphs over the various timeframes, or am I not understanding the way that RRDTool generates these graphs? I did notice that if I zoom in from one of the higher level graphs to match a more granular graph, I get the same numbers that I expect in the more granular graph. For example, in my example above, if I go to the 12-day graph with spikes at 30% and zoom in to just the last 48 hours, the graph looks the same as the 48-hour graph with the spikes at 60%.
Thanks, Ray
This communication is the property of Qwest and may contain confidential or privileged information. Unauthorized use of this communication is strictly prohibited and may be unlawful. If you have received this communication in error, please immediately notify the sender by reply e-mail and destroy all copies of the communication and any attachments.
It's because at longer history the values are averaged over longer time intervals. Steve
On Tue, Oct 26, 2010 at 3:07 PM, Lee, Raymond <Raymond.Lee at qwest.com> wrote:
Hi all,
When I go to the trend page for a host and then click on a graph to get the 48-hr, 12-day, 48-day, and 576-day graphs, something doesn't look right.
For example, I'm looking at the CPU utilization graphs:
- In the 48-hour graph, I see 2 spikes for "user" this morning that were near 60%.
- In the 12-day graph, those same 2 spikes only go up to about 30%.
- In the 48-day graph, the spikes only go up to about 15%.
- In the 576-day graph, the values hover around 10%.
Furthermore, the legend at the bottom of the graphs shows max values that are lower than what can be seen on the graphs. I'm seeing the same behavior on the other graphs from the trend page...not just the CPU Utilization graphs.
Is there a problem with the scaling of the y-axis between the graphs over the various timeframes, or am I not understanding the way that RRDTool generates these graphs? I did notice that if I zoom in from one of the higher level graphs to match a more granular graph, I get the same numbers that I expect in the more granular graph. For example, in my example above, if I go to the 12-day graph with spikes at 30% and zoom in to just the last 48 hours, the graph looks the same as the 48-hour graph with the spikes at 60%.
Thanks, Ray
This communication is the property of Qwest and may contain confidential or privileged information. Unauthorized use of this communication is strictly prohibited and may be unlawful. If you have received this communication in error, please immediately notify the sender by reply e-mail and destroy all copies of the communication and any attachments.
-- The test of a democracy is not the magnificence of buildings or the speed of automobiles or the efficiency of air transportation, but rather the care given to the welfare of all the people. -Helen Adams Keller, lecturer and author (1880-1968)
Truth never damages a cause that is just. -Mohandas Karamchand Gandhi (1869-1948)
From: sholmes42 at gmail.com [mailto:sholmes42 at gmail.com] On Behalf Of Steve Holmes Sent: Tuesday, October 26, 2010 3:21 PM To: xymon at xymon.com Subject: Re: [xymon] Question about trend graph scaling
It's because at longer history the values are averaged over longer time intervals. Steve
It would seem that is what is happening, but is that appropriate for things like CPU values? 100% of the CPU is constant over time. And it looks like when he takes a certain path to zoom in, he sees expected values.
On Tue, Oct 26, 2010 at 3:07 PM, Lee, Raymond <Raymond.Lee at qwest.com> wrote: Hi all, When I go to the trend page for a host and then click on a graph to get the 48-hr, 12-day, 48-day, and 576-day graphs, something > > doesn't look right. For example, I'm looking at the CPU utilization graphs:
- In the 48-hour graph, I see 2 spikes for "user" this morning that were near 60%.
- In the 12-day graph, those same 2 spikes only go up to about 30%.
- In the 48-day graph, the spikes only go up to about 15%.
- In the 576-day graph, the values hover around 10%.
On Tue, Oct 26, 2010 at 3:26 PM, Rob Munsch <Munsch at phillycarshare.org>wrote:
From: sholmes42 at gmail.com [mailto:sholmes42 at gmail.com] On Behalf Of Steve Holmes Sent: Tuesday, October 26, 2010 3:21 PM To: xymon at xymon.com Subject: Re: [xymon] Question about trend graph scaling
It's because at longer history the values are averaged over longer time intervals. Steve
It would seem that is what is happening, but is that appropriate for things like CPU values? 100% of the CPU is constant over time. And it looks like when he takes a certain path to zoom in, he sees expected values.
I don't know about anyone else but that is exactly the behavior I expect. If the CPU value stays at 100 for long enough it will show up at 100 on all the graphs. But if it peaks at 100 for 10 minutes and then goes back to .1 where it normally sits, then the average over, for example, 5 hours including that 10 minutes is well below 100. The if you zoom in on that same 5 hours you can see the details within that time span including the peak. Where's the problem?
Steve
On Tue, Oct 26, 2010 at 3:07 PM, Lee, Raymond <Raymond.Lee at qwest.com> wrote: Hi all,
When I go to the trend page for a host and then click on a graph to get the 48-hr, 12-day, 48-day, and 576-day graphs, something > > doesn't look right.
For example, I'm looking at the CPU utilization graphs:
- In the 48-hour graph, I see 2 spikes for "user" this morning that were near 60%.
- In the 12-day graph, those same 2 spikes only go up to about 30%.
- In the 48-day graph, the spikes only go up to about 15%.
- In the 576-day graph, the values hover around 10%.
-- The test of a democracy is not the magnificence of buildings or the speed of automobiles or the efficiency of air transportation, but rather the care given to the welfare of all the people. -Helen Adams Keller, lecturer and author (1880-1968)
Truth never damages a cause that is just. -Mohandas Karamchand Gandhi (1869-1948)
In <AANLkTi=UHCRZ=zYJWktNTCdMbTonh92W5NvL2JNzgpF2 at mail.gmail.com> Steve Holmes <sholmes42 at mac.com> writes:
It's because at longer history the values are averaged over longer time intervals.
It would seem that is what is happening, but is that appropriate for things like CPU values? 100% of the CPU is constant over time. And it looks like when he takes a certain path to zoom in, he sees expected values.
When you look at a graph, rrdtool will automatically decide which of the 4 datasets it is going to use - the 48-hour set, the 12-day set etc. - based on what timeperiod you are looking at. So if you start with the 12-day set but zoom into a period within the past 48-hours, then it will automatically switch from using the 12-day averaged set to the higher-precision 48-hour set - that's why you then see the peak just like you do on the 48-hour graph.
Wait a couple of days, and when you zoom in on the same peak, it won't be as high as it was today.
Regards, Henrik
-----Original Message----- From: Henrik Størner [mailto:henrik at hswn.dk] Sent: Tuesday, October 26, 2010 4:14 PM To: xymon at xymon.com Subject: Re: [xymon] Question about trend graph scaling
In <AANLkTi=UHCRZ=zYJWktNTCdMbTonh92W5NvL2JNzgpF2 at mail.gmail.com> Steve Holmes <sholmes42 at mac.com> writes:
It's because at longer history the values are averaged over longer time intervals.
It would seem that is what is happening, but is that appropriate for things like CPU values? 100% of the CPU is constant over time. And it looks like when he takes a certain path to zoom in, he sees expected values.
When you look at a graph, rrdtool will automatically decide which of the 4 datasets it is going to use - the 48-hour set, the 12-day set etc. - based on what timeperiod you are looking at. So if you start with the 12-day set but zoom into a period within the past 48-hours, then it will automatically switch from using the 12-day averaged set to the higher-precision 48-hour set - that's why you then see the peak just like you do on the 48-hour graph.
Wait a couple of days, and when you zoom in on the same peak, it won't be as high as it was today.
Ah, so the different graphs are averaged over time. I was confused because there are min., max., and avg. values in the legend for the CPU Util. graphs. So I guess the avg. there is really an average of averages. Thanks everyone!
Ray
Regards, Henrik
To unsubscribe from the xymon list, send an e-mail to xymon-unsubscribe at xymon.com
This communication is the property of Qwest and may contain confidential or privileged information. Unauthorized use of this communication is strictly prohibited and may be unlawful. If you have received this communication in error, please immediately notify the sender by reply e-mail and destroy all copies of the communication and any attachments.
participants (4)
-
henrik@hswn.dk
-
Munsch@phillycarshare.org
-
Raymond.Lee@qwest.com
-
sholmes42@mac.com