Using Traceroute to Measure Network Latency and Packet Loss

In this article, we explain the main concepts of Traceroute and how it can be used to measure network latency and packet loss. We also address the main limitations of such a tool that makes it unfit for accurate measurements.

What do you use Traceroute for?

Traceroute is a very useful tool for network monitoring and problems diagnostics. With traceroute you can:

• discover the network path between a source and a destination
• measure the network latency to reach each hop on the path
• measure the packet loss at each hop

It is extremely useful to measure the network quality (congestions, …). It can also detect any network path variation that may occur through BGP peering/routing configuration changes.

How does Traceroute work?

The main principle of traceroute is shown hereunder:

The whole traceroute principle is based on the TTL (Time To Live) field of the IP packet header. This field is mainly used to avoid loops in networks, where packets could be routed indefinitely in a loop under certain circumstances. When a host sends a packet on a network, its initial TTL value is between generally 32 and 255, depending on the operating system used. Each time the packet hits a router and must be routed, the TTL value decreases by 1. When a packet with a TTL value of 1 hits a router, this latter cannot route the packet anymore (it would mean a TTL=0). So it drops the packet (to avoid potential looping problem). It also informs the source about this by sending a specific ICMP error message back to the source (ICMP error message “TTL Exceeded In Transit”).

So traceroute relies on the fact that routers will send this ICMP error message back to the source in case of a TTL value reaching 1. When performing a traceroute, this is what happens:

1. The source sends a first IP packet with a TTL header field value of 1.
2. The packet hits the first router in the network path. The router drops the packet due to this TTL value and sends an ICMP error message back to the source.
3. The source has discovered the first router! So let’s go to the next one. For this, it sends a packet with a TTL value of 2.
4. The packet hits the first router, is routed normally, and the TTL values is decreased by 1 (new TTL value = 1).
5. The packet hits the second router in the path. The router drops it due to the TTL value of 1. Again an ICMP error message is sent back to the source, which discovers the second router in the path.
6. … and so on up until the packet reaches the final destination.
7. The type of message this final destination will send to the source depends on the specific traceroute implementation used (see next topic).

Different traceroute flavors

With traceroute, the IP packets are not sent as is. They are typically transported in a upper transport layer protocol like UDP, or directly in ICMP packets.

Traceroute in Windows

The standard Windows implementation of traceroute uses ICMP as protocol to send IP packets.

Let’s see how it works in practice:

These pictures show a Windows traceroute to google.com.

One line corresponds to a discovered router. By default, a Windows traceroute sends three packets per hop. You can see this through the 3 network latency values provided per line.

In this example, the packet has reached the destination after 7 consecutive hops. The last line (8) corresponds to the final destination itself.

Let’s now see in details what happened by using Wireshark. We analyze the fifth discovered router.

The left side screenshot shows that the source (192.168.1.31) sends an ICMP “Echo (ping) request” packet to the destination (216.58.208.110) with an IP header TTL field value of 5.

The right side screenshot shows the answer from the discovered intermediate router (91.183.245.122). This latter sends an ICMP error message “Time-to-live exceeded” back to the source. As additional data, it also sends back the packet that triggered this error message.

When the packet finally hits the destination, the packet does not have to be routed anymore. So the destination does not send any ICMP error message back to the source. This is shown hereunder.

At the left side, you can see that the source still sends ICMP packets with incremental IP field TTL value (8 in this case). Nevertheless, as the destination does not have to route the packet anymore, it does not care about the TTL value. Instead, it responds to the ICMP Echo request solicitation by sending an ICMP Echo reply back to the source.

Traceroute in Linux

The standard Linux traceroute uses UDP as transport protocol to send IP packets.

Let’s see how it works in practice:

Without looking at more details, this looks exactly the same as a Windows traceroute: 3 tests per TTL value providing the network latency to each routing hop, up to the destination.

Let’s see what exactly happened this time:

At the left side, you see the details of the fifth line of the traceroute, that is a UDP packet sent with an IP header TTL field of 5. At the right side, you see that an intermediate router sends an ICMP error message back (“TTL exceeded in transit”) to the source. You can check the reason of this ICMP error message from this ICMP packet to be sure this message has been triggered by the initial UDP packet.

When the packet finally hits the destination, the packet does not have to be routed anymore. So the destination does not send any ICMP error message back to the source. Nevertheless, as Linux uses UDP by default and not ICMP like Windows, the destination will not answer with an ICMP Echo reply back to the source. Instead, it answers by letting the source know it does not listen to the destination UDP port.

As you can see on the screenshot above, the final destination (172.217.168.206) sends an ICMP error message “Destination unreachable – Port unreachable” back to the source. It informs the source it does not listen to the destination UDP port (33.468 in our case).

Generally, traceroute based on UDP transport protocol uses destination ports above 33.435.

Main challenges/limitations with traceroute

If you want to quickly discover a route to a destination and have a first idea of the network performance, traceroute is certainly one of your best friends. Nevertheless, you should be aware of some major limitations that we can categorize as follows:

• nodes discovery
• path discovery
• performance metrics

Network nodes discovery

In order to identify all intermediate nodes up to the destination, they should send ICMP error messages back to the source. If one of them does not respond, it will not be identified. You can still know that there is a router in the path, but cannot clearly identify it.

On the screenshot above, the fourth hop router is not identified. This is symbolized by the * .

Lot of network nodes will not respond to ICMP solicitation for security reason. A firewall may simply filter some traffic without sending ICMP packet back to hide its presence.

Network path discovery

As you know by now, traceroute is based on packet probes with incremental IP header TTL field value. So the following scenario may perfectly happen:

It shows why traditional traceroute does not handle load balanced network traffic correctly. Instead of d