Contents

Interface MTU Introduction

The MTU (Maximum Transfer Unit) is the maximum packet size that can be transmitted across an interface. The MTU of an interface depends on the type of interface. This section only applies to Ethernet interfaces. The MTU value includes the IP and additional (e.g., TCP) protocol headers.

TCP/IP example for an Ethernet interface for which the default MTU is 1500:

The maximum size of the payload data varies depending on the transport layer protocol used. It can also be smaller as in the example above, for example, if a VPN is used which requires its own protocol informatin within the packet. In case of TCP/IP, the actual maximum payload size is also called MSS (Maximum Segment Size).

Fragmentation

The IP protocol includes the option to split up data packets into smaller packets (fragments) that are too big for the MTU of an interface. The recipient of the packet is responsible for reassembling the packets.

Possible Problems with IPv4 Fragmentation

The need to fragment IP packets can cause several problems. Some common problems are described below.

Performance Problems

Excessive fragmentation and reassembly causes additional network load due to more packets being transmitted. It also causes additional CPU load on systems that are required to fragment and reassemble packets. In addition, the loss of one fragment causes the retransmission of the whole packet. So in case of network instability, additional load is added to an already unstable network connection.

Connectivity Problems

If there are intermediate systems between two communication partners that have smaller MTUs than the two communication partners, data transmission may fail, for example in the following cases:

• If the intermediate systems do not allow fragmentation or block packet fragments.
• If the intermediate systems block the ICMP messages necessary for Path MTU Discovery.

For a TCP connections, the two communication partners come to an agreement regarding the MSS to be used based on their own interface MTU. If any intermediate systems have a smaller MTU but either the two communication partners don't know this (failed Path MTU Discovery) and the intermediate systems cannot fragment the data packets, any packet that is too large will be dropped.

For protocols without the ability to negotiate the payload size or cannot handle fragmented traffic, data will be lost as soon as a packet is larger than the smallest MTU on the path.

Example with two ICMP data sizes across a DSL connection that is limited to a MTU of 1492. The IP + ICMP headers are 28 bytes in total. In the example, fragmentation is disabled (-M do). The source system has an MTU of 1500 bytes.

The first attempt with an attempted size of 1493 bytes fails, the second with a packet size of 1492 succeeds.

Implications for Charon Emulators and the Legacy Guest Systems

When dedicating a Charon host NIC to the emulator for use by the legacy guest system, ensure that the MTU of the dedicated NIC is not smaller than the MTU interface visible in the guest system.

Typically, legacy guest systems, such as Solaris, HP-UX, Tru64, and OpenVMS have a default MTU of 1500 bytes.

Physical Charon hosts usually also have a default MTU of 1500 bytes.

However, in cloud environments, the default MTU of Ethernet interfaces configured for a Charon host can vary. The following table provides an overview:

Caveat: ensuring that the Charon host MTU is not smaller than the guest system MTU prevents problems arising at the emulator host. However, it cannot prevent problems arising from smaller MTU sizes along the complete communication path if fragmentation does not work properly. Therefore, it is recommended to identify the actual MTU, for example, of the path across a VPN between cloud and on-premises networks and to either ensure that fragmentation and path discovery work correctly, or adapt the MTU of the legacy guest system accordingly.