Search: Advanced search
Browse by category:
Identifying and Resolving Packet Loss
When everything is working properly, each data packet should reach its destination. However, when there are problems with links or routers (e.g. overload), packets may be lost. It takes considerable time to detect and resend a lost packet, so performance drops dramatically with even a relatively small amount of packet loss. As a rule of thumb, packet loss should be less than 1 percent. Packet loss of more than 5 percent is serious.
Can you get to your service provider? PING your DNS server.
Internet providers supply at least one, and usually two, DNS servers (which convert names into actual Internet addresses) to subscribers. These are "nearby" nodes that usually respond to PING -- a good way to test your basic connection. Call your provider to request the numbered IP address of the DNS server. PING the numbered IP address at least 10 times. You should get consistently low latency with no packet loss, as in the example above. If you don't, run a traceroute/tracert to pinpoint the problem.
If you're on a cable modem, your local subnet may be overloaded, particularly if you see this problem predominately at times of peak demand. If you're on DSL, then your DSL modem may be having problems with synch or errors. Either problem needs to be addressed by your provider.
Problem 1B: Your provider's network or DNS server
The DNS server or your provider's internal network may be overloaded or malfunctioning. These are serious problems that can be fixed only by your provider.
Example of problem 1B
Can you get outside of your provider? Traceroute to a remote node.
Problem 2A: DNS failure
You may have the name wrong. Check it. Also check other names. If the name is valid, particularly if other names are also failing, either your Internet configuration is screwed up (e.g. DNS servers are not being configured correctly) or the provider's DNS server is acting up. You may be able to fix a configuration problem yourself (e.g. by manual configuration of DNS servers), but a DNS server problem is a serious issue that can be fixed only by your provider.
Example of problem 2A -- deliberately bogus name
Problem 2B: Internet transit overload
Most providers are connected to the Internet by means of "transit" links, typically a small number of high-speed circuits that each carry traffic from a large number of subscribers. If a transit link gets saturated, which can easily happen if a provider isn't proactive about keeping capacity ahead of subscriber traffic growth, latency will climb at the provider's "border" router (node) as packets are queued to wait their turn. If the border router runs out of queuing capacity, it copes by simply discarding packets (packet loss).
The solution is for your provider to install more capacity or for you to find a better provider if the problem isn't fixed right quick! In the following example, note the big jump in latency at the transition between the dummy ISP (example.net) and the transit. Each "*" is a lost packet. What matters is the first node at which the problem shows up.
Example of problem 2B
Problem 2C: Internet overload
Capacity problems on Internet "backbone" circuits and major connection nodes are much less common than some providers would have us believe. Further, most providers will claim they're not at fault when it happens. However, they do sometimes occur, particularly when there's a critical failure (e.g. major cable inadvertently cut by a clumsy backhoe operator). This would be much like the example immediately above, except that the problem shows up at a later node.
Problem 2D: Routing problem
When routing "tables" (the Internet "map") get messed up, packets may get sent to the wrong place or even stuck in a loop. There usually isn't much your provider can do about this except to report the problem. Note how the following example gets stuck in a loop (that continues until the packet expires) between ccc and ddd.
Example of problem 2D
Basic measurement tools (software that's probably already on your computer)
Traces the path between you and the remote host as a series of nodes (and the links between them). Typically displays three latency measurements for each node and looks up node names (if available). Under Windows it's called "tracert," as in the examples below. If you use Linux, substitute "traceroute" for "tracert." Here's an example of a clean, fast DSL connection: TP
Tracing route to www.techtv.com [18.104.22.168] over a maximum of 30 hops:
1 21 ms 17 ms 15 ms gateway.example.net
Sends a special packet designed to get a response back from a particular remote node, much like the echo of a sonar ping used to detect objects underwater. It's usually possible to send several (or even continuous) PINGs in succession, and the latency of each PING is commonly displayed. Since PING can be used in certain forms of Internet attacks, some nodes deliberately ignore PINGs as a security measure, so a complete failure to respond to PING is not necessarily bad. Note the consistent latency and lack of packet loss in this example of a clean, fast DSL connection (where "-n 10" is used to send 10 PINGs):
ping -n 10 www.TechTV.com