where does xymon client get tcp packet data?
The below relates to linux, xymon 4.3.17
I was going to write a script to collect and report numbers of TCP packet retransmissions. But when I look at the 'trends' for this host, I see there is already a chart for "TCP/IP statistics" which includes "In/Out/Retrans" I'm specifically looking for TCP retransmission count. This counter does not seem to line up with information being reported by nstat.
What is this built-in test reporting, and from where is it collected?
Do things because you should, not just because you can.
John Thurston 907-465-8591 John.Thurston at alaska.gov Enterprise Technology Services Department of Administration State of Alaska
On 1/13/2016 11:21 AM, John Thurston wrote:
The below relates to linux, xymon 4.3.17
I was going to write a script to collect and report numbers of TCP packet retransmissions. But when I look at the 'trends' for this host, I see there is already a chart for "TCP/IP statistics" which includes "In/Out/Retrans" - snip -
Ohhh. I think I see. It probably _is_ lining up with the numbers being reported by nstat. The difficulty is the number is being graphed as packets/second and plotted on the same chart as IN and OUT packets (also being graphed in packets/second). My small number of retransmissions is being lost in the noise of the total volume of packets.
Is there a way to re-define the graph to display a multiple of the recorded value? Or plot this counter against an alternate y-axis set to a different scale?
I suppose I could define a new graph to contain only retransmissions so the small number isn't swamped by the general traffic count.
-- Do things because you should, not just because you can.
John Thurston 907-465-8591 John.Thurston at alaska.gov Enterprise Technology Services Department of Administration State of Alaska
On 1/13/2016 12:38 PM, John Thurston wrote:
On 1/13/2016 11:21 AM, John Thurston wrote:
The below relates to linux, xymon 4.3.17
I was going to write a script to collect and report numbers of TCP packet retransmissions. But when I look at the 'trends' for this host, I see there is already a chart for "TCP/IP statistics" which includes "In/Out/Retrans" - snip -
Ohhh. I think I see. It probably _is_ lining up with the numbers being reported by nstat. The difficulty is the number is being graphed as packets/second and plotted on the same chart as IN and OUT packets (also being graphed in packets/second). My small number of retransmissions is being lost in the noise of the total volume of packets.
Yep. It varies by client OS, but the interface and network statistics are coming from the ifconfig and netstat commands, or their local equivalents. Those sections in the raw client message are isolated and turned into distinct data messages by xymond_client, which are interpreted and stored by xymond_rrd. It's the only two-level message re-injection occurring out of the box.
Is there a way to re-define the graph to display a multiple of the recorded value? Or plot this counter against an alternate y-axis set to a different scale?
I suppose I could define a new graph to contain only retransmissions so the small number isn't swamped by the general traffic count.
This would probably be your best option. As long as it's coming out of a single RRD file, you can slice and dice the individual RRA's up in various ways at once for different views.
-jc
On Jan 13, 2016, at 15:38, John Thurston <john.thurston at alaska.gov> wrote:
On 1/13/2016 11:21 AM, John Thurston wrote:
The below relates to linux, xymon 4.3.17
I was going to write a script to collect and report numbers of TCP packet retransmissions. But when I look at the 'trends' for this host, I see there is already a chart for "TCP/IP statistics" which includes "In/Out/Retrans" - snip -
Ohhh. I think I see. It probably _is_ lining up with the numbers being reported by nstat. The difficulty is the number is being graphed as packets/second and plotted on the same chart as IN and OUT packets (also being graphed in packets/second). My small number of retransmissions is being lost in the noise of the total volume of packets.
Is there a way to re-define the graph to display a multiple of the recorded value? Or plot this counter against an alternate y-axis set to a different scale?
I suppose I could define a new graph to contain only retransmissions so the small number isn't swamped by the general traffic count.
-- Do things because you should, not just because you can.
Maybe I'm missing the point, but unless one has a strictly two-host link, full-duplex, both similar (speed, OS), and lightly loaded, retransmissions are _normal_ at a low level. And if the level is so low relative to traffic that it's invisible, why is the exact level interesting? Should it become visible, there's still time to worry.
Your signature slogan made me think about that. :-) Or is this one of those government someone-had-an-idea-that-you-can't-escape things (been there)?
participants (3)
-
cleaver@terabithia.org
-
john.thurston@alaska.gov
-
rlhamil2@gmail.com