Skip to content

Transit Gateway

AWS Transit Gateway connects VPCs, VPN connections and Direct Connect gateways through a single regional hub. It behaves like a cloud router: each network is attached once, and the transit gateway’s route tables decide what can reach what. That replaces the mesh of point-to-point peering connections that otherwise grows quadratically with the number of VPCs.

Hub-and-spoke diagram: four Amazon VPCs, a VPC VPN connection to a customer gateway, and a Direct Connect gateway all attached to a single AWS Transit Gateway at the centre

  • Transitive routing between attached networks. Unlike VPC peering, traffic can pass through the hub from one attachment to another.
  • Hub-and-spoke by design; segmentation comes from using multiple transit gateway route tables rather than from separate gateways.
  • Regional. Transit gateways in different Regions are joined with peering attachments.
  • Shareable across accounts with AWS Resource Access Manager, so one network account can own the gateway and member accounts attach to it.
  • Works with Site-to-Site VPN and with Direct Connect through a Direct Connect gateway and a transit virtual interface.
  • Supports IP multicast, which VPC peering and Direct Connect do not.
  • Supports Transit Gateway Connect attachments, which carry GRE tunnels to third-party SD-WAN appliances at higher bandwidth than a VPN attachment.
QuotaDefaultAdjustable
Transit gateways per account per Region5Yes
Attachments per transit gateway5,000Yes
Transit gateways a single VPC can attach to5No
Peering attachments per transit gateway50Yes
Route tables per transit gateway20Yes
Total routes across all route tables on one transit gateway10,000On request
Bandwidth per VPC attachment per Availability ZoneUp to 100 Gbps each directionOn request
Bandwidth per Connect peer (GRE tunnel)Up to 5 GbpsNo
Direct Connect gateways per transit gateway20No
Transit gateways per Direct Connect gateway6No

The MTU is 8,500 bytes for VPC, Direct Connect, Connect and peering attachments, and 1,500 bytes over VPN.

Note the first row in particular: several transit gateways per Region is normal, and separating production from non-production is a common reason to have more than one.

A transit VPC builds the same hub from EC2 instances running third-party routing or VPN software. It predates Transit Gateway and is now the fallback rather than the default.

Transit VPCTransit Gateway
OperationYou run, patch and monitor the appliance instancesFully managed
ScalingBounded by the instance types you choseScales without intervention
BandwidthInstance-dependentUp to 100 Gbps per VPC attachment per Availability Zone
RoutingManaged on the appliancesTransit gateway route tables, with multiple tables for segmentation
CostEC2 instances plus vendor licences plus dataAttachment-hours plus data processed
Cross-RegionBuilt by the appliancesPeering attachments
MulticastDepends on the applianceSupported natively

A transit VPC is still defensible when you need a specific vendor feature — a particular firewall’s inspection or routing policy engine — or when you are heavily invested in an existing deployment that works. For new designs, Transit Gateway is the default.

Note that inspection appliances no longer require a transit VPC either: a Gateway Load Balancer puts third-party appliances in a traffic path without making them the router.

Site-to-Site VPN. Terminate an IPsec VPN on the transit gateway from Google Cloud, Microsoft Azure, Oracle Cloud or IBM Cloud. Straightforward, and bounded by VPN tunnel bandwidth unless you aggregate tunnels with ECMP.

Transit Gateway Connect. Attach a third-party SD-WAN or virtual router appliance over GRE, with higher bandwidth and lower latency than a VPN attachment, and BGP for routing.

Direct Connect. Where a partner has presence in more than one cloud, a Direct Connect circuit into a transit gateway gives the most predictable performance.

Whichever you choose, plan the address space first — overlapping private ranges between clouds cannot be routed — and expect cross-Region traffic and inter-cloud data transfer to appear on the bill.

Multicast sends one copy of a stream that the network replicates only where it is needed, instead of one copy per recipient (unicast) or a copy to everybody on the segment (broadcast). Streaming 1,000 receivers costs one stream from the source rather than 1,000.

  • Multicast group addresses are 224.0.0.0 to 239.255.255.255 for IPv4.
  • Receivers join and leave groups with IGMP.
  • Routers build distribution trees so traffic only reaches segments with interested receivers.

Typical uses are live video distribution, market data feeds, and software distribution to many hosts at once.

Transit gateway multicast has its own limits — throughput per flow is capped at 1 Gbps and aggregate throughput per Availability Zone at 20 Gbps — so it is not a fit for high-frequency trading or other latency-critical distribution. Check the multicast quotas before committing to a design.