Maximum transmission unit

The limit on the maximum frame size is imposed for several reasons:

• To reduce retransmission time in the event of packet loss or irreparable distortion. The probability of loss increases with increasing packet length.
• In order for the host to not occupy the channel for a long time during half-duplex operation (the Interframe gap is also used for this purpose).
• The larger the packet being sent, the longer the other packets will wait, especially in serial interfaces . Therefore, the small MTU was relevant in times of slow dial-up connections .
• Small size and speed of network buffers for incoming and outgoing packets. However, too large buffers also degrade performance .

The MTU value is determined by the standard of the corresponding protocol, but can be redefined automatically for a specific stream (PMTUD protocol) or manually for the desired interface. On some interfaces, the default MTU may be set below the maximum possible.

The MTU value is limited from below, usually the minimum allowable frame length. The minimum frame size is usually associated with a collision domain (half duplex channel). Domain members must learn about the collision before the end of the frame transfer. It also takes into account the restrictions imposed by the standard on the physical features of the tract, for example, length, materials.

Minimum frame size by standards:
Fast Ethernet 100Base-T: 64 bytes
Gigabit Ethernet 1000Base-T: 512 bytes

For a high-performance network, the reasons behind the initial MTU restrictions are outdated. In this regard, the standard for Jumbo frames with an increased MTU was developed for Ethernet.

Some protocols, for example from the ATM protocol group, operate on fixed-length cells. There is no way for them to change the MTU. In some cases, this approach leads to better management of network flows.

Nominal rate - bit rate of data transmission supported on the transmission interval of one packet without distinguishing service and user headers. Obviously, the smaller the packet, the faster it will be transmitted.
Effective speed is the average transmission rate of user data (load) in a stream. This parameter depends on the ratio of service header lengths to load. In many cases, a certain amount of information is more efficiently transmitted in large packets due to lower overhead. Correctly comparing the flow rates is possible only for the same amount of data that you want to transfer at a time.

The greater the payload of a packet with respect to its headers, the higher the rate of transmission of useful information and the percentage of channel utilization. Since MTU is the upper limit of the payload length, maximum speed is achieved when the loads of all packets in the stream have a length equal to MTU. For shaping, there are different methods for calculating the Ethernet frame length (without Ethernet frame ) (without jumbo):

• up to 1514 bytes (1518 with the 802.1Q tag) on ​​general-purpose computers, where the shaper is implemented programmatically. Final bytes (called or footer) FCS fields are processed by the network adapter and a packet without them is transmitted to the operating system.
• up to 1518 bytes (1522 with 802.1Q tag) - Ethernet packet, including the FCS field, but not considering the preamble.
• up to 1538 bytes (1542 with the 802.1Q tag) on ​​special network platforms. ^[2] Together with the usual Ethernet header, the preamble, FCS, and frame interval are taken into account. This method of calculation should also be used when the speed is limited only by the transmission protocol (in this case Ethernet) and is not limited by the provider.

There are several ways to increase throughput in the context of packet lengths:

• Reducing the frame interval. For Fast Ethernet, it defaults to 0.96 μs for which 12 bytes could be transmitted. Some full-duplex network adapters can transmit and receive frames with reduced frame spacing.
• Header Compression Implementation Example - Jacobson TCP / IP Header Compression Method .
• Low level compression An example implementation is IP Payload Compression Protocol .
• The solution to the problem of transmitting small data (tinygram problem) by keeping small data in the buffer to send one large packet. An example implementation is the Nagle algorithm . A special case of this problem is “chatty” protocols that send a lot of requests where one can get by. There are software and hardware solutions that optimize such connections, for example, at .

The host knows the MTU value for its own (and possibly its neighbors ) interface, but the minimum MTU value for all network nodes is usually unknown. Another potential problem is that higher layer protocols can create larger packets that are not supported by other network nodes.

To resolve these problems, IP supports fragmentation , which allows you to break the datagram into smaller pieces, each of which will be small enough to freely pass through the node, due to which fragmentation occurs. Packet fragments are marked so that the IP of the target host can assemble the fragments into the original datagram. Packet fragmentation has its drawbacks:

• The target host must receive all the fragments in order to assemble the packet. If at least one fragment is lost, the entire packet must be resent.
• As a rule, fragmentation is performed by slower processors than those that process transit traffic. In some cases, delays caused by fragmentation of all packets may be deemed unacceptable.
• The load on the processors of intermediate routers and destination machines is increasing.
• The memory requirement of the router increases, as it is necessary to store all fragments of the IP packet in the buffer.
• Some firewalls may block fragments.
• Due to the copy of the IP header, bandwidth is inefficiently used for each fragment.
• Routing protocols, such as OSPF , require a consistent MTU to work correctly.

The following terms are used in Ethernet networks to designate frames that are of a non-standard size:

• RUNT - (shorty) - a frame whose size is less than 64 bytes (512 bits).
• LONG - (long) - a frame whose size lies in the range from 1518 to 6000 bytes.
• GIANT - (giant) - a frame whose size exceeds 6000 bytes.

The term Path MTU means the smallest MTU in the path of a packet on the network.

Although fragmentation solves the problem of incompatibility of packet sizes and MTU values, it significantly reduces the performance of network devices. In this regard, an alternative technology called Path MTU discovery ( RFC 1191 ) was proposed in 1988. The essence of the technology is that when connecting two hosts, the DF (don't fragment) parameter is set, which prevents packet fragmentation. This causes the node, whose MTU is less than the size of the packet, to reject the transmission of the packet and send the ICMP message "fragmentation is necessary, but its prohibition flag (DF) is set." The sending host reduces the packet size and resends it. Such an operation occurs until the packet is small enough to reach the destination host without fragmentation.

However, this technology also has potential problems. Some routers are configured by administrators to completely block ICMP packets (this is not very competent, but may be the easiest solution to several security problems). As a result, if the packet size does not correspond to the MTU value in a certain section, the packet is discarded, and the sender host cannot receive information about the MTU value and does not resend the packet. Therefore, the connection between the hosts is not established. The problem was called MTU Discovery Black Hole ( RFC 2923 ), and the protocol was modified to detect such routers. One of the common problems when working on a network that blocks ICMP packets is working in IRC , when the user successfully logs in, but cannot receive the so-called MOTD (message of the day), as a result of which it is not possible to use the network.

The problem poses a potential danger to any PPPoE connection using MTU less than typical (1500 bytes).

There are several solutions to this problem. The simplest, of course, is to cancel the filtering of ICMP packets. However, often such an operation is beyond the user's competence. Therefore, the problem is solved by manually adjusting the size of the transmitted packet on the user gateway. To do this, change the MSS value (the maximum segment size, that is, a value less than the MTU by 40 bytes in the case of IPv4 ). When a connection is established, hosts exchange information about the maximum size of a segment that each of them can accept. Therefore, by changing the MSS value, both hosts are forced to exchange packets, which the user gateway will be able to accept without fragmentation.

In addition, there are methods to cancel the DF bit in order to resolve fragmentation. However, the MSS adjustment method is preferred.

On Cisco routers, the tunnel PMTUD implementation only supports TCP packets. ^[3]

• Maximum Receive Unit (MRU) - maximum size of the useful PPP data block.
• Maximum segment size (MSS) - The maximum size of the useful data block of the TCP segment.
• (PDU) - general protocol data unit (payload + headers). An alternative not always correct definition is the entire data block of a certain level of the network model.
• (SDU) - in OSI terminology, a data unit that has passed from its level to the lower level of the model and has not yet been encapsulated in PDUs of this level.

• Marc Slemko. Path MTU Discovery and ICMP Filtering (Neopr.) ( January 18, 1998 ). Date of treatment March 31, 2010. Archived August 26, 2011.
• DrTCP - Utility for optimizing MTU in Microsoft Windows
• mturoute - console utility for troubleshooting MTU
• Manual setting of MTU values ​​in the registry of Windows Mobile, Windows XP, Windows Vista (unspecified) (January 21, 2009). Archived on August 26, 2011.
• explanation of MTU of various levels on the example of mikrotik os
• Resolve IP Fragmentation, MTU, MSS, and PMTUD Issues with GRE and IPSEC