Since the internet connection in our house does break down from time to time I set up a little experiment:

For the last two month, one of my machines is pinging google.com on an half-hourly basis. One measurement consists of 50 pings.

I now calculated the mean percentage of packets lost for each hour of the day: percent of lost packets

My questions:

  1. Could this peak in the evening be caused by choosing google.com as the ping destination?
  2. Would you recommend using an other destination and which?
  3. Does this indicate that something is wrong with my connection?
  4. What would be a better strategy to measure where exactly the problem in our internet connection is? Our ISP tells us it is working fine so I try to aggregate some proof...


Edit: I forgot to mention that the machine is directly connected to the router (no WiFi). And the router is pinged as well, with no packet loss at all.

  • What exactly do you mean by " internet connection in our house does break down from time to time"? If "break down" means "stop working", tracking packet loss when it does work is unlikely to tell you anything useful. – Isaac Rabinovitch Mar 12 '13 at 23:52
  • That's right, but my interest is in when does it break down, and how often/how long. – Dirk Mar 13 '13 at 9:25

Unfortunately you really have not provided enough information to work out where the problem is. To answer as best as I possibly can with the limited information provided:

  1. If my experiences are anything to go by, pinging Google is normally quite a good bet, as they design their network to be as fast as possible. Also as ICMP is prioritized, evening peak probably does not make a significant difference - particularly in terms of packet loss - which I'd argue should be 0.

  2. Google is a good destination, but to get a better idea of what is going on you may want to additionally try pinging your gateway, and if they allow it, your providers DNS, mail or web server. This will help show where the packet loss is creeping in. Realistically though, at the level of packet loss you are seeing, look at downloading MTR (or WinMTR) and running that at peek to get a better idea of where packet loss is coming in.

  3. Subjectively, 5% packet loss is at the top end of acceptable for a Wifi based network - assuming you are not saturating your network. On the flip side, I get upset about 0.5% packet loss on my fibre connections - as a point of reference, loosely speaking, less then 1% is OK for VOIP, above
    that not so much. If you expect to be able to use Skype or Viber or what have you on your connection then 5% packet loss is not OK. For just Web browsing it may just about suffice.

  4. As an ISP, I want to see the results of an MTR, which show the latencies and packet loss between the destination - this helps me look where the bottleneck might be and is a good first step. I would also like to know when the test was done so I can correlate it with the customers other usage and what is going on on the system. The packet loss graphs you have done are useful as well, but not in isolation.

    As a client, my ISP has been unable to excuse away my graphs, which plot packet loss (I do it for 250 pings, once a second over 5 minute intervals, combined with minimimum, average and maximum latencies for those pings). I also have a set of graphs showing my utilization of the link, and have sets of graphs showing local (ie very close to me), and to another POP they own of specific interest a few hundred KM's away.

Other observations:

It looks like your latency increases in the afternoons - which means the first places I would be looking are if the problem is WIFI when everyone around me is using it. After ruling that out I'd start questioning my ISP about oversubscribing the connection.

  • Thanks a lot for your answers. The machine is directly connected to the router, and it pings the router as well which shows no package los at all. MTR seems to be what I was looking for. – Dirk Mar 7 '13 at 1:07

This more than likely is a result of congestion somewhere along the line. It could be your router but more likely an upstream provider.

You don't state how you're doing the 50 pings e.g. what time interval, are you waiting for one to fail/succeed before the next or firing 50 off all at once (flood pinging).

Such a loss during periods of high congestion is not unusual in my experience. It can be down to lower prioritisation for ICMP traffic but more likely is happening for the same percentage of all connections - it's just TCP will gracefully resend and reorder packets so you are less likely to notice.

To gain a better picture of the situation I would suggest you implement the following:

  1. Increase the interval between your pings
  2. Ping an IP address for google not the domain - google.com will return a number of A records and it may be you're using different end IPs (and hence routing differently) without knowing it
  3. Record the mean time to respond; see if this correlates with loss - if it does so you see higher ping roundtrip times and higher loss then it indicates congestion. You could then investigate by storing traceroute logs instead and see if there is a likely bottleneck somewhere where you're seeing suddenly increased times
  4. Try pinging more than google. When I've benchmarked network performance in the past I've done it using 4 or 5 good end-points (again with the IP address not the hostname) so you can rule out congestion or a specific problem within just google's network causing you to question your entire connection

This is typical of most residential ISP accounts. You're seeing a peak due to network congestion as people get home after work, then go on-line for the whole evening. This sort of evening peak is especially pronounced in high-tech communities with lots of on-line gamers (like where I live, here in Redmond, home of Microsoft.)

Not the answer you're looking for? Browse other questions tagged or ask your own question.