Niel bad day at the ranch above. Jerry nice and clean in TX, just a few TCP/deep ICMP timeouts.
FIRSTLY, note the left hand scale. Originally it varied between plots, but I have changed it to vary only if the script detects timeouts. It will be 4000 for curl, 1900 for dns query timeouts, 1000 for only ICMP timeouts. At times, the ISP will neglect to respond to ICMP pings under congestion periods, so you would see a straight line across at 1000 msec.
The “far end (deep) Customer ICMP ping” timeout can be much greater since it represents points far away from the customer [other customers].
SECONDLY, the daily plot -received in the email- STARTS at
2AM [yesterday] for all pings, goes to the right, and wraps to the
left at 1AM, representing 1AM this morning. There is always a gap
from 1AM to 2AM. When the plot is called from the rpi web page, it
will start at 2AM and end at the current time.
The plot uses “lines” to connect the points for the
near, deep ICMP and the TCP pings, and you will see straight
connecting lines when the plotting is discontiguous. The “far end
(deep)” Customer ICMP pings only happen about 50 min for every
hour, the others are pretty much every minute of each hour [except
for 1AM to 2AM]. The “offline,” “tcp offline,” "google
curl” and "dns query" plots just plot the points.
The “offline” points represent when I have detected offline by detecting a timeout for the tcp syn ping, and simultaneously detecting a near end ICMP ping timeout.
The “tcp offline” points represent when any one of the top100Web did not respond within a second. It is not unusual to see some of these in a plot. The far end (deep) Customer ICMP ping timeouts are plotted for each timeout.
The “google curl” points are the delay to pull down the index page for the configured curl target [www.google.com is the default] [none of the accompanying images and pulled]. These “curl” draws only occur twice an hour, bracketing the deep ICMP pings. The "dns query time" plot actually times a dns query for www.google.com. This will be using your router (which is tested for DNS query) which will be caching the IP address, so it can often be taken as a "router response time," tho not always.
If you have an rpi, you can go to the rpi web page [linked in your daily email] and invoke a current plot. This is an especially nice feature and lets you see more detail since it is for less than the 24 hours reported in the email nightly plot. It takes about a minute for the plot to show up, so you can refresh until you see the current time in the plot [it will show the last current plot until it is updated]. Only the current time is shown on the plot, not the date. You can also invoke a "ping burst" where 60 ICMP ping request responses to the near ICMP target is plotted.
Entering the first link, you will see 3 plots for each customer. The daily delay plot showing all 3 plots with overlaying lines representing offline and tcp timeouts, plus the offline and google points. The “archive” plot is an archive which plots the average of each of the first three plots every day. The speedtest archive records the speedtest results, up and down for each customer.
The sample plot below are representative of errored links -the current ones are in the first link.
Take a look at the second link to see the range of technologies represented, from fiber, cable, VDSL, ADSL, WADSL. I even had a Hughes sat at one time.
Niel
bad day at the ranch above. Jerry nice and clean in TX, just a few
TCP/deep ICMP timeouts.
