Edit

Use Source Network Address Translation (SNAT) for outbound connections

  • Article

Certain scenarios require virtual machines or compute instances to have outbound connectivity to the internet. The frontend IPs of a public load balancer can be used to provide outbound connectivity to the internet for backend instances. This configuration uses source network address translation (SNAT) to translate virtual machine's private IP into the load balancer's public IP address. SNAT maps the IP address of the backend to the public IP address of your load balancer. SNAT prevents outside sources from having a direct address to the backend instances.

Azure's outbound connectivity methods

The following methods are Azure's most commonly used methods to enable outbound connectivity:

# Method Type of port allocation Production-grade? Rating
1 Use the frontend IP address(es) of a load balancer for outbound via outbound rules Static, explicit Yes, but not at scale OK
2 Associate a NAT gateway to the subnet Dynamic, explicit Yes Best
3 Assign a public IP to the virtual machine Static, explicit Yes OK
4 Default outbound access use Implicit No Worst

Diagram of Azure outbound options.

1. Use the frontend IP address of a load balancer for outbound via outbound rules

Diagram public load balancer with outbound rules.

Outbound rules enable you to explicitly define SNAT (source network address translation) for a standard SKU public load balancer. This configuration allows you to use the public IP or IPs of your load balancer for outbound connectivity of the backend instances.

This configuration enables:

  • IP masquerading

  • Simplifying your allowlists

  • Reduces the number of public IP resources for deployment

With outbound rules, you have full declarative control over outbound internet connectivity. Outbound rules allow you to scale and tune this ability to your specific needs via manual port allocation. Manually allocating SNAT port based on the backend pool size and number of frontendIPConfigurations can help avoid SNAT exhaustion.

You can manually allocate SNAT ports either by "ports per instance" or "maximum number of backend instances". If you have virtual machines in the backend, it's recommended that you allocate ports by "ports per instance" to get maximum SNAT port usage.

Calculate ports per instance as follows:

Number of frontend IPs * 64K / Number of backend instances

If you have Virtual Machine Scale Sets in the backend, it's recommended to allocate ports by "maximum number of backend instances". If more VMs are added to the backend than remaining SNAT ports allowed, scale out of Virtual Machine Scale Sets could be blocked, or the new VMs won't receive sufficient SNAT ports.

For more information about outbound rules, see Outbound rules.

2. Associate a NAT gateway to the subnet

Diagram of a NAT gateway and public load balancer.

Azure NAT Gateway simplifies outbound-only Internet connectivity for virtual networks. When configured on a subnet, all outbound connectivity uses your specified static public IP addresses. Outbound connectivity is possible without load balancer or public IP addresses directly attached to virtual machines. NAT Gateway is fully managed and highly resilient.

Using a NAT gateway is the best method for outbound connectivity. A NAT gateway is highly extensible, reliable, and doesn't have the same concerns of SNAT port exhaustion.

For more information about Azure NAT Gateway, see What is Azure NAT Gateway.

3. Assign a public IP to the virtual machine

Diagram of virtual machines with instance level public IP addresses.

Associations Method IP protocols
Public IP on VM's NIC SNAT (Source Network Address Translation)
isn't used.		 TCP (Transmission Control Protocol)
UDP (User Datagram Protocol)
ICMP (Internet Control Message Protocol)
ESP (Encapsulating Security Payload)

Traffic returns to the requesting client from the virtual machine's public IP address (Instance Level IP).

Azure uses the public IP assigned to the IP configuration of the instance's NIC for all outbound flows. The instance has all ephemeral ports available. It doesn't matter whether the VM is load balanced or not. This scenario takes precedence over the others.

A public IP assigned to a VM is a 1:1 relationship (rather than 1: many) and implemented as a stateless 1:1 NAT.

4. Default outbound access

Diagram of default outbound access.

Note

This method is NOT recommended for production workloads as it adds risk of exhausting ports. Please refrain from using this method for production workloads to avoid potential connection failures.

Default outbound access is when An Azure resource is allocated a minimal number of ports for outbound. This access occurs when the resource meets any of the following conditions:

  • doesn't have a public IP associated to it.
  • doesn't have a load balancer with outbound Rules in front of it.
  • isn't part of Virtual Machine Scale Sets flexible orchestration mode.
  • doesn't have a NAT gateway resource associated to its subnet.

Some other examples of default outbound access are:

  • Use of a basic SKU load balancer
  • A virtual machine in Azure (without the associations mentioned above). In this case, outbound connectivity is provided by the default outbound access IP. This IP is a dynamic IP assigned by Azure that you can't control. Default SNAT isn't recommended for production workloads and can cause connectivity failures.
  • A virtual machine in the backend pool of a load balancer without outbound rules. As a result, you use the frontend IP address of a load balancer for outbound and inbound and are more prone to connectivity failures from SNAT port exhaustion.

What are SNAT ports?

Ports are used to generate unique identifiers used to maintain distinct flows. The internet uses a five-tuple to provide this distinction.

If a port is used for inbound connections, it has a listener for inbound connection requests on that port. That port can't be used for outbound connections. To establish an outbound connection, an ephemeral port is used to provide the destination with a port on which to communicate and maintain a distinct traffic flow. When these ephemeral ports are used for SNAT, they're called SNAT ports.

By definition, every IP address has 65,535 ports. Each port can either be used for inbound or outbound connections for TCP (Transmission Control Protocol) and UDP (User Datagram Protocol). When a public IP address is added as a frontend IP to a load balancer, 64,000 ports are eligible for SNAT. While all public IPs that are added as frontend IPs can be allocated, frontend IPs are consumed one at a time. For example, if two backend instances are allocated 64,000 ports each, with access to two frontend IPs, both backend instances consume ports from the first frontend IP until all 64,000 ports have been exhausted.

Each port used in a load balancing or inbound NAT rule consumes a range of eight ports from the 64,000 available SNAT ports. This usage reduces the number of ports eligible for SNAT, if the same frontend IP is used for outbound connectivity. If load-balancing or inbound NAT rules consumed ports are in the same block of eight ports consumed by another rule, the rules don't require extra ports.

How does default SNAT work?

When a VM creates an outbound flow, Azure translates the source IP address to an ephemeral IP address. This translation is done via SNAT.

If using SNAT without outbound rules via a public load balancer, SNAT ports are pre-allocated as described in the following default SNAT ports allocation table:

Default port allocation table

The following table shows the SNAT port preallocations for backend pool sizes:

Pool size (VM instances) Default SNAT ports per IP configuration
1-50 1,024
51-100 512
101-200 256
201-400 128
401-800 64
801-1,000 32

Port exhaustion

Every connection to the same destination IP and destination port uses a SNAT port. This connection maintains a distinct traffic flow from the backend instance or client to a server. This process gives the server a distinct port on which to address traffic. Without this process, the client machine is unaware of which flow a packet is part of.

Imagine having multiple browsers going to https://www.microsoft.com, which is:

  • Destination IP = 23.53.254.142

  • Destination Port = 443

  • Protocol = TCP

Without SNAT ports for the return traffic, the client has no way to separate one query result from another.

Outbound connections can burst. A backend instance can be allocated insufficient ports. Use connection reuse functionality within your application. Without connection reuse, the risk of SNAT port exhaustion is increased.

For more information about connection pooling with Azure App Service, see Troubleshooting intermittent outbound connection errors in Azure App Service

New outbound connections to a destination IP fail when port exhaustion occurs. Connections succeed when a port becomes available. This exhaustion occurs when the 64,000 ports from an IP address are spread thin across many backend instances. For guidance on mitigation of SNAT port exhaustion, see the troubleshooting guide.

For TCP connections, the load balancer uses a single SNAT port for every destination IP and port. This multiuse enables multiple connections to the same destination IP with the same SNAT port. This multiuse is limited if the connection isn't to different destination ports.

For UDP connections, the load balancer uses a port-restricted cone NAT algorithm, which consumes one SNAT port per destination IP whatever the destination port.

A port is reused for an unlimited number of connections. The port is only reused if the destination IP or port is different.

Constraints

  • When a connection is idle with no new packets being sent, the ports will be released after 4 – 120 minutes.

  • This threshold can be configured via outbound rules.

  • Each IP address provides 64,000 ports that can be used for SNAT.

  • Each port can be used for both TCP and UDP connections to a destination IP address

  • A UDP SNAT port is needed whether the destination port is unique or not. For every UDP connection to a destination IP, one UDP SNAT port is used.

  • A TCP SNAT port can be used for multiple connections to the same destination IP provided the destination ports are different.

  • SNAT exhaustion occurs when a backend instance runs out of given SNAT Ports. A load balancer can still have unused SNAT ports. If a backend instance’s used SNAT ports exceed its given SNAT ports, it's unable to establish new outbound connections.

  • Fragmented packets are dropped unless outbound is through an instance level public IP on the VM's NIC.

  • Secondary IP configurations of a network interface don't provide outbound communication (unless a public IP is associated to it) via a load balancer.

Next steps