what is mtu - Elastic Compute Service - Alibaba Cloud Documentation Center

Tune the Maximum transmission unit (MTU) to help reduce fragmentation, lower latency, and improve throughput across your network path.

MTU classifications

MTU types vary by network layer and use case.

Link-layer MTU
The maximum data payload size at the data link layer. In Ethernet, the MTU is typically 1500 bytes (excluding the Ethernet header). A standard Ethernet frame is 1518 bytes: destination MAC (6 bytes) + source MAC (6 bytes) + Type/Length (2 bytes) + Data payload (1500 bytes) + frame check sequence (4 bytes).
IP MTU
The largest packet the network layer can process, including the IP header and data payload but excluding the Ethernet frame header and trailer. In Alibaba Cloud, MTU usually refers to IP MTU.
With minimum IP and TCP header sizes (which can grow with options), the 1500-byte IP MTU breaks down to: IP header (20 bytes) + TCP header (20 bytes) + Data payload (1460 bytes).
Jumbo frame MTU
Jumbo frames expand the standard Ethernet frame to 9 KB. A jumbo frame includes a Layer 2 Ethernet header and a CRC field, which are not part of the IP MTU. Jumbo frames reduce packet count and improve efficiency. Alibaba Cloud supports jumbo frames up to 8500 bytes. See Jumbo frames.
Path Maximum Transmission Unit (PMTU)
The PMTU is the largest packet size that can traverse the entire path between two nodes without fragmentation. The path includes all devices and links (routers, vSwitches, communication links) from source to destination.
PMTUD dynamically discovers the PMTU on a path to avoid fragmentation. See Path MTU Discovery (PMTUD).

Factors that affect the MTU size

Several factors determine the MTU size.

Network devices: Routers and vSwitches have maximum MTU values determined by their hardware. The smallest MTU on a path limits all communication on that path.
Network type: Different network types have different MTU requirements. Wireless networks often use a smaller MTU because signal instability increases the risk of packet loss with large packets.
Transmission medium: Twisted-pair cables, optical fibers, and wireless media may impose different MTU limits.
Network protocol: Protocols define their own MTU. The standard Ethernet MTU is 1500 bytes.
Application requirements: Specific applications may require a non-standard MTU, which affects the overall network MTU.

How the MTU affects network performance

An incorrect MTU affects performance in the following ways:

Fragmentation and reassembly
When a packet crosses a network boundary with a smaller MTU, it splits into smaller fragments. These fragments must be reassembled at the destination. This adds latency, reduces efficiency, and increases processing load.
Reduced transmission efficiency
Fragmentation overhead lowers overall throughput. A lost fragment can incorrectly trigger TCP retransmission, further degrading performance.
Network congestion
A too-small MTU creates excessive fragments, increasing congestion. A larger MTU reduces packet count, decreasing congestion and improving throughput.

Set a consistent MTU across all devices on the path, as large as possible without exceeding any link's limit. This minimizes fragmentation and improves efficiency. For high-bandwidth scenarios such as video conferencing or large file transfers, tuning the MTU improves transmission quality.

View the MTU of an ECS network interface

All ECS instances default to an MTU of 1500 bytes. Instances with jumbo frames enabled support 8500 bytes. Connect to your instance and run the following commands to check the current NIC MTU.

View the MTU of a Windows instance

Sample operating system: Windows Server 2022.

Sample instance family: ecs.g8i.large.

Connect to the ECS instance.
See Connect to a Windows instance using Workbench.
Open Command Prompt or Windows PowerShell.

View the MTU of the current network interface:

netsh interface ipv4 show subinterfaces

Before you enable jumbo frames

The MTU of the primary NIC named Ethernet and the secondary Elastic Network Interface (ENI) named Ethernet 2 is 1500, which is the default value.

C:\Windows\system32>netsh interface ipv4 show subinterfaces

   MTU  MediaSenseState   Bytes In    Bytes Out  Interface
------  ---------------  ----------  ----------  -------------
4294967295            1          0           0  Loopback Pseudo-Interface 1
      1500            1   15133287     1860371  Ethernet
      1500            1       1519        3234  Ethernet 2

After you enable jumbo frames

The MTU of the primary NIC named Ethernet and the secondary ENI named Ethernet 2 is 8500.

C:\Users\Administrator>netsh interface ipv4 show subinterfaces

   MTU  MediaSenseState   Bytes In   Bytes Out  Interface
------  ---------------  ---------  ---------  -------------
4294967295                1          0          0  Loopback Pseudo-Interface 1
  8500                1      75314    1062294  Ethernet
  8500                1    2386860     631043  Ethernet 2

View the MTU of a Linux instance

Sample operating system: Alibaba Cloud Linux 3.2.

Sample instance family: ecs.g8i.large.

Connect to the ECS instance.
See Connect to a Linux instance using Workbench.

View the MTU of the current network interface:

ip a

Before you enable jumbo frames

The MTU of the primary NIC eth0 and the secondary ENI eth1 is 1500, which is the default value.

[root@xxx ~]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:14:4d:09 brd ff:ff:ff:ff:ff:ff
    altname enp0s5
    altname ens5
    inet 172.17.0.139/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359999sec preferred_lft 315359999sec
    inet6 fe80::216:3eff:fe14:4d09/64 scope link
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:3f:f1:43 brd ff:ff:ff:ff:ff:ff
    altname enp0s7
    altname ens7
    inet 172.17.0.138/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359999sec preferred_lft 315359999sec
    inet6 fe80::24c1:8e90:7d8a:fc43/64 scope link noprefixroute
       valid_lft forever preferred_lft forever

After you enable jumbo frames

The MTU of the primary NIC eth0 and the secondary ENI eth1 is 8500.

[root@xxx ~]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 8500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:14:4d:09 brd ff:ff:ff:ff:ff:ff
    altname enp0s5
    altname ens5
    inet 172.17.0.139/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359996sec preferred_lft 315359996sec
    inet6 fe80::216:3eff:fe14:4d09/64 scope link
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 8500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:3f:f1:43 brd ff:ff:ff:ff:ff:ff
    altname enp0s7
    altname ens7
    inet 172.17.0.138/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359996sec preferred_lft 315359996sec
    inet6 fe80::24c1:8e90:7d8a:fc43/64 scope link noprefixroute
       valid_lft forever preferred_lft forever

Use the ping command to manually probe the PMTU

Manually probe the PMTU with the ping command:

Warning

This test may affect network communication. Perform it during off-peak hours.

Test method: Send a packet with the Don't Fragment (DF) flag set. Gradually increase the packet size until you receive an ICMP "Fragmentation needed and DF set" error. The packet size just before the error equals the PMTU minus the IP header (20 bytes) and ICMP header (8 bytes).
Test command: Test the MTU size of the network link:
```
ping -c 1 -M do -s [packet_size] [destination_address]
```
Parameters:
- -c 1: Send one packet.
- -M do: Set the Don't Fragment (DF) flag.
- -s [packet_size]: The ICMP PDU size. Start with a common value such as 1472.
  Note
  The ping command uses ICMP over IP. The ICMP header is 8 bytes and the IP header is 20 bytes. So -s 1472 produces a total IP packet of 20 + 8 + 1472 = 1500 bytes.
Test example:
- MTU information for hosts at both ends:
  FROM: 172.17.0.127. Jumbo frames are enabled on the ECS instance (MTU = 8500 bytes). See Enable or disable jumbo frames.
  TO: 172.17.0.128. The MTU is 1500.
- Test: Send a packet of 8472 bytes (excluding IP and ICMP headers):
  ping -c 1 -M do -s 8472 172.17.0.128
- Result: An ICMP error "Frag needed and DF set(mtu=1500)" is returned, indicating the packet exceeds the path MTU (1500 bytes). Because the DF flag is set, forwarding devices cannot fragment the packet, so it is dropped.
```
[root@          ~]# ping -c 1 -M do -s 8472 172.17.0.128
PING 172.17.0.128 (172.17.0.128) 8472(8500) bytes of data.
From 172.17.0.253 icmp_seq=1 Frag needed and DF set (mtu = 1500)

--- 172.17.0.128 ping statistics ---
1 packets transmitted, 0 received, +1 errors, 100% packet loss, time 0ms
```

Manually change the MTU of a network interface

Some networking services, such as VPN Gateway and traffic mirroring, require smaller packets from ECS instances. Change the ECS NIC MTU to meet these requirements.

Warning

Before you change the MTU, understand how a different MTU might affect network performance. An improper value can cause fragmentation and reduce efficiency.
- The Linux MTU is obtained from the DHCP server automatically. If DHCP is active and you are not in a tunneling scenario (such as VPN Gateway), you typically do not need to manually change the NIC MTU.
- For scenarios such as jumbo frames, follow the relevant best practices instead of manually changing the MTU. See Enable or disable jumbo frames.
Ensure the new MTU does not exceed the instance's maximum supported MTU. The maximum is 8500 bytes with jumbo frames enabled, or 1500 bytes when jumbo frames are disabled or unsupported.

The following example changes the MTU of an ECS NIC to 1399 bytes.

Windows instance

Sample operating system: Windows Server 2022.

Connect to the Windows instance.
See Connect to a Windows instance using Workbench.
Open Command Prompt or Windows PowerShell.

View the MTU of the Ethernet NIC:

netsh interface ipv4 show subinterfaces

The response shows the current MTU of the Ethernet NIC is 1500.

C:\Users\Administrator>netsh interface ipv4 show subinterfaces

   MTU  MediaSenseState   Bytes In    Bytes Out  Interface
------  ---------------  ----------  ----------  -------------
4294967295                1          0           0  Loopback Pseudo-Interface 1
      1500                1   76093074     3922261  Ethernet
      1500                1        943        3068  Ethernet 2

Details of the netsh interface ipv4 show subinterfaces command

Run the following command to view parameter details.

netsh interface ipv4 show subinterfaces ?

The command returns the following output, which includes the usage, parameters, and description.

C:\Users\Administrator>netsh interface ipv4 show interfaces ?

Usage: show interfaces [[interface=]<string>] [[rr=]<integer>]
        [[level=]normal|verbose] [[store=]active|persistent]

Parameters:

      Tag             Value
      interface - Interface name or index.
      rr        - Refresh rate for interface statistics (in seconds).
      level     - One of the following values:
                  normal: Display one line per interface. This is the
                          default when no interface is specified.
                  verbose: Display additional information for each
                           interface. This is the default when an
                           interface is specified.
      store     - One of the following values:
                  active: Show interfaces in the stack (default).
                  persistent: Show persistent interfaces.

Description: Display information about all interfaces, or if an
      interface is specified, display information about the
      specified interface.

Set the MTU of the Ethernet interface to 1399 permanently:

netsh interface ipv4 set subinterface "Ethernet" mtu=1399 store=persistent

"Ok." confirms the command succeeded. If an error is returned, resolve it as prompted.

Verify the MTU of the Ethernet NIC. The change has taken effect.

C:\Users\Administrator>netsh interface ipv4 show subinterfaces

   MTU  MediaSenseState  Bytes In  Bytes Out  Interface
------  ---------------  ---------  ---------  -------------
4294967295                1          0          0  Loopback Pseudo-Interface 1
      1399                1   76123457    4064932  Ethernet
      1500                1       1099       3556  Ethernet 2

Details of the netsh interface ipv4 set subinterface command

Run the following command to view parameter details.

netsh interface ipv4 set subinterface ?

The command returns the following output, which includes the usage, parameters, and description.

C:\Users\Administrator>netsh interface ipv4 set subinterface  ?

Usage: set subinterface [interface=]<string> [[mtu=]<integer>]
        [[subinterface=]<string>] [[store=]active|persistent]

Parameters:

       Tag            Value
       interface  - Interface name or index.
       mtu        - MTU of the subinterface. The default is the natural
                    MTU of the link.
       subinterface - Subinterface LUID. Required only for interfaces
                      with multiple subinterfaces.
       store      - One of the following values:
                    active: Persist setting only until the next boot.
                    persistent: Persist setting permanently.

Description: Modify subinterface configuration parameters.

Example:

       set subinterface "1" mtu=1500 store=active

Linux instance

Sample operating system: Alibaba Cloud Linux 3.2.

Temporarily change the MTU using the ip command

Important

This temporarily changes the NIC MTU. Restarting the network service or instance reverts the change.

Connect to the Linux instance.
See Connect to a Linux instance using Workbench.

View the current MTU of the NIC:

ip a

The primary NIC eth0 and secondary ENI eth1 both have an MTU of 1500.

[root@xxx ~]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:0e:de:c3 brd ff:ff:ff:ff:ff:ff
    altname enp1s3
    inet 172.17.0.127/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359332sec preferred_lft 315359332sec
    inet6 fe80::216:3eff:fe0e:dec3/64 scope link
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:37:eb:bf brd ff:ff:ff:ff:ff:ff
    altname enp1s5
    inet 172.17.0.122/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359332sec preferred_lft 315359332sec
    inet6 fe80::216:3eff:fe37:ebbf/64 scope link
       valid_lft forever preferred_lft forever

Temporarily change the MTU of eth1 to 1399:

sudo ip link set dev eth1 mtu 1399

Verify the NIC information. The MTU of eth1 is now 1399.

[root@xxx ~]# sudo ip link set dev eth1 mtu 1399
[root@xxx ~]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:de:de:c3 brd ff:ff:ff:ff:ff:ff
    altname enp1s3
    inet 172.17.0.127/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359127sec preferred_lft 315359127sec
    inet6 fe80::216:3eff:fede:dec3/64 scope link
       valid_lft forever preferred_lft forever
3: eth1  <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1399 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:37:eb:bf brd ff:ff:ff:ff:ff:ff
    altname enp1s5
    inet 172.17.0.122/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359127sec preferred_lft 315359127sec
    inet6 fe80::216:3eff:fe37:ebbf/64 scope link
       valid_lft forever preferred_lft forever

Edit the network configuration file to make the change permanent

Important

Edit the network configuration file to permanently change the MTU. Restart the network service or instance to apply.
The network configuration file varies based on the Linux distribution and version. In the following example, the Alibaba Cloud Linux 3.2 image, which is compatible with Red Hat Enterprise Linux (RHEL) and CentOS, is used.

If you attached a secondary ENI, such as eth1, and need to change its MTU, ensure the ENI is correctly configured. See Configure a secondary ENI.

Connect to the Linux instance.
See Connect to a Linux instance using Workbench.

View the current MTU of the NIC:

ip a

The MTU of the primary NIC eth0 and the secondary ENI eth1 is 1500.

[root@ ~]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:0e:de:c3 brd ff:ff:ff:ff:ff:ff
    altname enp1s3
    inet 172.17.0.127/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359332sec preferred_lft 315359332sec
    inet6 fe80::216:3eff:fe0e:dec3/64 scope link
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:37:eb:bf brd ff:ff:ff:ff:ff:ff
    altname enp1s5
    inet 172.17.0.122/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359332sec preferred_lft 315359332sec
    inet6 fe80::216:3eff:fe37:ebbf/64 scope link
       valid_lft forever preferred_lft forever

View the network interface configuration files:

ls /etc/sysconfig/network-scripts

The response shows the configuration files for eth0 and eth1.

[root@xxx ~]# cd /etc/sysconfig/network-scripts/
[root@xxx network-scripts]# ls -l
total 8
-rw-r--r-- 1 root root 170 Sep 19 10:11 ifcfg-eth0
-rw-r--r-- 1 root root 179 Sep 19 10:43 ifcfg-eth1

Disable the network configuration feature of cloud-init:

sudo vim /etc/cloud/cloud.cfg.d/99-disable-network-config.cfg

Add the following line:

network: {config: disabled}

Edit the network configuration file of eth1 and add the MTU setting:

sudo vi /etc/sysconfig/network-scripts/ifcfg-eth1

Example file content after editing (add MTU=1399 at the end of the file):

# Created by cloud-init on instance boot automatically, do not edit.
#
BOOTPROTO=dhcp
DEVICE=eth1
HWADDR=00:16:3e:37:eb:bf
METRIC=200
ONBOOT=yes
TYPE=Ethernet
USERCTL=no
MTU=1399

Restart the network service to apply the configuration:

sudo systemctl restart NetworkManager

Verify the network interface information. The MTU of eth1 is now 1399.

[root@xxx network-scripts]# sudo systemctl restart NetworkManager
[root@xxx network-scripts]# ip a
1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
    link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
    inet 127.0.0.1/8 scope host lo
       valid_lft forever preferred_lft forever
    inet6 ::1/128 scope host
       valid_lft forever preferred_lft forever
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:0e:de:c3 brd ff:ff:ff:ff:ff:ff
    altname enp1s3
    inet 172.17.0.127/24 brd 172.17.0.255 scope global dynamic noprefixroute eth0
       valid_lft 315359946sec preferred_lft 315359946sec
    inet6 fe80::216:3eff:fe0e:dec3/64 scope link
       valid_lft forever preferred_lft forever
3: eth1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1399 qdisc mq state UP group default qlen 1000
    link/ether 00:16:3e:37:eb:bf brd ff:ff:ff:ff:ff:ff
    altname enp1s5
    inet 172.17.0.122/24 brd 172.17.0.255 scope global dynamic noprefixroute eth1
       valid_lft 315359946sec preferred_lft 315359946sec
    inet6 fe80::216:3eff:fe37:ebbf/64 scope link
       valid_lft forever preferred_lft forever

MTU limits in practice

MTU limits for ECS-to-ECS communication

When an ECS instance that supports jumbo frames is the source, the maximum link MTU to the destination varies by scenario.

Destination host	Maximum allowed MTU
An ECS instance with jumbo frames enabled in the same VPC or in a peered VPC in the same region	8500
An ECS instance without jumbo frames enabled in the same VPC or in a peered VPC in the same region	1500
Any ECS instance in a VPC connected through CEN
A host in an on-premises data center connected through Express Connect
An external host accessed through an EIP or public IP address

See VPC peering connections and VPC connections.

MTU limits of Alibaba Cloud networking services

Note the MTU limits of networking services:

Within the Alibaba Cloud network, links default to 1500 bytes MTU. When communicating with ECS instances that do not support jumbo frames, note the MTU limits of network components such as VPN Gateway.
When using non-TCP protocols (such as UDP or ICMP) to access networking services, note the MTU limits of those services.

Most Alibaba Cloud services do not support jumbo frames. The following table shows MTU limits for some networking services.

Network Products	Maximum supported MTU	Description
NAT Gateway	1500	None
Server Load Balancer (including ALB, NLB, and CLB)		ALB and CLB do not forward fragmented packets. When using UDP or ICMP, packets from the source ECS instance must not exceed 1500 bytes.
Alibaba Cloud services (such as OSS and RDS)		Traffic passes through Server Load Balancer products. Note their limits.
Global Accelerator		None
Cloud Firewall		None
PrivateLink		None
VPN Gateway	1399	See MTU configuration for an IPsec-VPN connection.

Best practices for setting the MTU

Follow these practices to reduce fragmentation, lower latency, and increase throughput:

Determine a suitable MTU:
- The default MTU for Ethernet is 1500 bytes. See MTU classifications.
- Probe the PMTU with the ping command. See Use the ping command to manually probe the PMTU.
Enable and adapt to PMTUD: Enable PMTUD on your OS and network devices so the path MTU is discovered dynamically. Applications then reduce packet size in response to ICMP messages, avoiding fragmentation. See How PMTUD works.
Consider network device limits: Match the MTU on routers, vSwitches, and other network devices to the OS MTU to prevent fragmentation or packet drops.
Adapt to specific applications: Real-time applications such as VoIP and video conferencing may benefit from a smaller MTU to reduce latency.
Regularly monitor network performance: Track throughput, latency, and application performance to adjust the MTU as network conditions change.