4.1 Azure ExpressRoute Architecture & Peering Models
Key Takeaways
- ExpressRoute Direct provides dual 10 Gbps or 100 Gbps physical fiber ports directly to Microsoft Enterprise Edge (MSEE) routers with support for line-rate MACsec (802.1AE) Layer 2 encryption.
- Private Peering connects directly to Azure Virtual Networks via RFC 1918 addresses, while Microsoft Peering connects to Azure PaaS and Microsoft 365 services via public IPs and requires Route Filters with BGP communities.
- FastPath bypasses the Virtual Network Gateway CPU data plane and programs routes directly into Azure host NICs, delivering line-rate throughput and lower latency on UltraPerformance and ErGw3AZ SKUs.
- ExpressRoute does not support jumbo frames: the documented maximum ExpressRoute MTU is 1,400 bytes, ExpressRoute gateways drop fragmented packets, and ExpressRoute Direct's 1,500-byte IP MTU is a physical port requirement, not an end-to-end payload size.
- The GatewaySubnet for ExpressRoute should be sized at /26 to ensure operational capacity for gateway maintenance, high-performance SKUs, and co-existence with Azure VPN Gateway.
Azure ExpressRoute creates a private, dedicated, high-throughput connection between enterprise on-premises infrastructure and Microsoft Cloud services. Unlike Site-to-Site VPN connections that traverse the public internet, ExpressRoute traffic travels over a private optical network fabric, delivering predictable performance, lower latency, and enhanced security for mission-critical enterprise workloads.
1. ExpressRoute Connectivity Models & Physical Topology
To establish an ExpressRoute circuit, organizations partner with network service providers or connect directly to Microsoft Enterprise Edge (MSEE) routers at designated ExpressRoute peering locations (meet-me facilities).
+---------------------------------------------------------------------------------------------------+
| EXPRESSROUTE CONNECTIVITY MODELS |
| |
| 1. Cloud Exchange Co-location (Layer 2 / Layer 3 Ethernet Exchange) |
| [ On-Premises Meet-Me ] <---> [ Cloud Exchange (e.g. Equinix, Megaport) ] <---> [ MSEE Pair ] |
| |
| 2. Point-to-Point Ethernet (Dedicated Layer 2 Connection) |
| [ Corporate Datacenter ] <============== Dedicated Point-to-Point Fiber ==============> [ MSEE Pair ] |
| |
| 3. Any-to-Any IPVPN (Layer 3 MPLS / IP-VPN Network) |
| [ Branch 1 ] ----+ |
| [ Branch 2 ] ----+---> [ Telco MPLS / WAN Provider Backbone ] <-----------------------> [ MSEE Pair ] |
| [ HQ Core ] ----+ |
| |
| 4. ExpressRoute Direct (Dedicated Dual 10G / 100G Direct Fiber Ports) |
| [ Customer Edge Router 1 ] <============== 10G/100G Direct Cross-Connect ==============> [ MSEE 1 ] |
| [ Customer Edge Router 2 ] <============== 10G/100G Direct Cross-Connect ==============> [ MSEE 2 ] |
+---------------------------------------------------------------------------------------------------+
Connectivity Models Breakdown
| Connectivity Model | Layer | Bandwidth Range | Best Use Case | Operational Characteristics |
|---|---|---|---|---|
| Cloud Exchange Co-location | Layer 2 / Layer 3 | 50 Mbps - 10 Gbps | Enterprises with infrastructure co-located in carrier-neutral facilities | Virtual cross-connects provisioned quickly via software-defined cloud exchanges (e.g., Equinix Cloud Exchange, Megaport). |
| Point-to-Point Ethernet | Layer 2 | 50 Mbps - 10 Gbps | Single primary datacenter requiring dedicated high-speed link | Dedicated point-to-point Ethernet circuits between on-premises sites and Azure meet-me facilities. |
| Any-to-Any (IPVPN / MPLS) | Layer 3 | 50 Mbps - 10 Gbps | Distributed branch offices already integrated into a managed WAN | Telecom provider integrates Azure as another branch node in the enterprise IPVPN cloud. Azure is managed like any other WAN site. |
| ExpressRoute Direct | Layer 1 / 2 | Dual 10 Gbps or Dual 100 Gbps | Massive data ingestion, financial trading, ultra-high throughput | Direct physical fiber connection to MSEE routers at peering locations. Supports active-active dual ports (up to 200 Gbps aggregate) and hardware MACsec encryption. |
ExpressRoute Direct Architecture & MACsec Encryption
ExpressRoute Direct provides direct physical connectivity into Microsoft's global backbone without an intermediary exchange provider. Key architectural capabilities include:
- Dual Physical Ports: Provisioned as active-active port pairs (e.g., $2 \times 100\text{ Gbps}$ or $2 \times 10\text{ Gbps}$) landing on distinct MSEE chassis in a specific peering location.
- Multi-Circuit Hosting: On a 100 Gbps ExpressRoute Direct port pair, administrators can carve out multiple isolated 1 Gbps, 2 Gbps, 5 Gbps, 10 Gbps, 40 Gbps, or 100 Gbps standard ExpressRoute circuits across different Azure subscriptions and routing domains.
- Line-Rate MACsec (802.1AE) Encryption: ExpressRoute Direct supports point-to-point Layer 2 hardware encryption between customer edge routers and MSEEs. Pre-shared keys (CAK/CKN) are securely stored in Azure Key Vault and rotated seamlessly without dropping BGP peering sessions.
2. ExpressRoute Peering Models: Private vs. Microsoft Peering
ExpressRoute supports two distinct BGP routing domains over a single physical circuit: Azure Private Peering and Microsoft Peering (Public Peering is legacy and deprecated).
+---------------------------------------------------------------------------------------------------+
| EXPRESSROUTE PEERING ARCHITECTURES |
| |
| +---------------------------------------------------------------------------------------------+ |
| | Microsoft Cloud | |
| | | |
| | +--------------------------------------------------------------------------------------+ | |
| | | Azure Private Peering (RFC 1918 / Private IPs) | | |
| | | -> Virtual Networks, Spoke VNets, Virtual Network Gateways, FastPath VM NICs | | |
| | +--------------------------------------------------------------------------------------+ | |
| | | |
| | +--------------------------------------------------------------------------------------+ | |
| | | Microsoft Peering (Public IPs / BGP Communities / Route Filters) | | |
| | | -> Microsoft 365 (Exchange, Teams, SharePoint) [Requires Authorization] | | |
| | | -> Azure PaaS (Storage, SQL DB, Key Vault, Azure App Services) [Public Endpoints] | | |
| | +--------------------------------------------------------------------------------------+ | |
| +----------------------------------------------+----------------------------------------------+ |
| | (Redundant BGP /30 Subnets) |
| v |
| +---------------------------------------------------------------------------------------------+ |
| | Microsoft Enterprise Edge (MSEE) | |
| +----------------------------------------------+----------------------------------------------+ |
| | (Primary & Secondary VLANs) |
| v |
| +---------------------------------------------------------------------------------------------+ |
| | On-Premises Customer Edge (CE) Routers | |
| | * NAT Gateway / Public IP Pool (Required for Microsoft Peering Outbound SNAT) | |
| | * Private IP Addressing (Private Peering) | |
| +---------------------------------------------------------------------------------------------+ |
+---------------------------------------------------------------------------------------------------+
Peering Comparison Matrix
| Architectural Attribute | Azure Private Peering | Microsoft Peering |
|---|---|---|
| Target Services | Virtual Networks (VNets), IaaS Virtual Machines, Azure Private Endpoints | Azure Public PaaS (Storage, SQL, Event Hubs) & Microsoft 365 SaaS |
| IP Addressing | RFC 1918 Private IPs (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) | Public IPv4 / IPv6 addresses owned by enterprise (registered in ARIN/RIPE/RADb) |
| Gateway Requirement | Requires Virtual Network Gateway (ExpressRoute SKU) in GatewaySubnet | No GatewaySubnet or Azure Gateway required (routes directly to PaaS/M365) |
| NAT Requirements | No NAT needed (direct private end-to-end IP routing) | Mandatory SNAT: On-premises private IPs must be NATted to registered Public IPs |
| Route Filtering | Azure automatically advertises VNet prefixes to on-premises | Route Filters must be attached to the circuit; BGP communities filter services |
| Microsoft Approval | No approval required (enabled immediately upon circuit provisioning) | Microsoft Authorization required for Microsoft 365 BGP community filters |
Microsoft Peering NAT & Route Filter Mechanics
- Public IP Validation: Public IP prefixes used for BGP peering and outbound NAT must be registered to the enterprise in a Regional Internet Registry (RIR) such as ARIN, RIPE, or APNIC, or validated through Routing Assets Database (RADb).
- NAT Architecture: Customer Edge routers must implement Source NAT (SNAT) so that internal client private IPs (e.g.,
10.1.0.0/16) are translated to public IPs before packets cross the Microsoft Peering BGP boundary. The NAT IP pool must not overlap with the /30 peering IP addresses. - Route Filters & BGP Communities: By default, an activated Microsoft Peering session receives zero routes. To receive BGP routes, administrators create an Azure Route Filter resource containing specific BGP community strings and link it to the ExpressRoute circuit:
12076:51004— Azure Global Services (Storage, SQL, Cosmos DB, Azure Monitor)12076:5010— Microsoft 365 Exchange Online (Requires Microsoft approval)12076:5020— Microsoft 365 SharePoint Online / OneDrive (Requires Microsoft approval)12076:5030— Microsoft 365 Teams / Skype for Business (Requires Microsoft approval)
3. ExpressRoute Virtual Network Gateways & SKUs
When connecting an Azure Virtual Network to an ExpressRoute circuit via Private Peering, you must deploy an ExpressRoute Virtual Network Gateway inside the dedicated GatewaySubnet.
ExpressRoute Gateway SKU Specifications
| Gateway SKU | Max Throughput (Standard Data Plane) | Max Circuit Connections | Zone Redundancy | FastPath Support |
|---|---|---|---|---|
| Standard | 1.0 Gbps | 4 circuits | No | No |
| HighPerformance | 2.0 Gbps | 8 circuits | No | No |
| UltraPerformance | 4.0 - 9.5 Gbps | 8 circuits | No | Yes |
| ErGw1AZ | 1.0 Gbps | 4 circuits | Yes (Zone-Redundant) | No |
| ErGw2AZ | 2.0 Gbps | 8 circuits | Yes (Zone-Redundant) | No |
| ErGw3AZ | 4.0 - 9.5 Gbps | 8 circuits | Yes (Zone-Redundant) | Yes (Full Feature Set) |
[!IMPORTANT] GatewaySubnet Sizing for ExpressRoute: The
GatewaySubnetmust be sized at/27minimum, but /26 is the recommended best practice. A/26subnet provides 59 usable host IPs, allowing concurrent deployment of an ExpressRoute Gateway, an Azure VPN Gateway (for co-existence failover), and necessary maintenance instances allocated during platform host updates.
4. ExpressRoute FastPath Architecture
In standard ExpressRoute Private Peering, all packets entering the Azure VNet from an on-premises network must pass through the virtual network gateway VM instances in the GatewaySubnet before reaching target virtual machines.
+-------------------------------------------------------------------------------------------------+
| STANDARD GATEWAY ROUTING vs. FASTPATH |
| |
| STANDARD ROUTING (Bottleneck at Gateway CPU): |
| [ On-Premises ] ---> [ MSEE ] ---> [ GatewaySubnet (VGW VMs: 1-9.5 Gbps) ] ---> [ VM NIC ] |
| |
| FASTPATH ROUTING (Bypasses Gateway CPU completely): |
| [ On-Premises ] ---> [ MSEE ] ================================================> [ VM NIC ] |
| (Direct Host SmartNIC Data Plane) |
| |
| * Note: ExpressRoute Gateway is still required for BGP Control Plane route exchange! |
+-------------------------------------------------------------------------------------------------+
FastPath Operational Mechanics
- Data Plane Bypass: When FastPath is enabled on the ExpressRoute Connection resource, Azure programs routing tables directly into the host hypervisors and Azure SmartNICs. Packets travel directly between the MSEE router and the destination VM NIC at full wire speed.
- Throughput Scaling: FastPath removes the gateway CPU bottleneck, scaling network throughput to 10 Gbps, 40 Gbps, and 100 Gbps when paired with ExpressRoute Direct.
- Latency Reduction: Bypassing gateway VM hops reduces inter-packet jitter and improves round-trip latency for high-frequency trading and big-data analytics.
- Prerequisites & SKUs: FastPath requires an UltraPerformance or ErGw3AZ (or ErGw1AZ/2AZ for standard FastPath configurations) gateway SKU.
FastPath Feature Support & Limitations
- Virtual Network Peering Transit: FastPath supports peered VNets. Traffic from on-premises to a peered spoke VNet bypasses the gateway completely.
- Private Endpoints: FastPath routes traffic directly to Azure Private Endpoints deployed in the VNet without gateway throttling.
- Internal Load Balancers (ILB): Supports backend pools behind ILBs across all SKUs.
- Control Plane Dependency: The ExpressRoute Gateway cannot be deleted; it is strictly required to maintain eBGP neighbor peering sessions with the MSEEs.
- UDR Bypassing: By default, User-Defined Routes (UDRs) associated with the
GatewaySubnetare bypassed by FastPath. To force FastPath traffic through an NVA/Firewall, specific UDR-on-GatewaySubnet FastPath support must be explicitly enabled.
5. MTU, Packet Size & Fragmentation
ExpressRoute does not support jumbo frames. This is one of the most commonly mis-remembered ExpressRoute facts, and it is worth memorising precisely because plausible-sounding 9,000-byte answers appear as distractors.
| Setting | Documented value | Where it applies |
|---|---|---|
| Maximum ExpressRoute MTU | 1,400 bytes | The ExpressRoute FAQ states ExpressRoute supports an MTU of 1,400 bytes. Tune VM TCP/IP settings against this number, not against 1,500. |
| Maximum TCP/UDP packet size through an ExpressRoute gateway | 1,400 bytes | Applies to traffic traversing the ExpressRoute virtual network gateway. |
| ExpressRoute Direct physical interface IP MTU | 1,500 bytes | A hardware onboarding requirement of the customer router/switch ports facing the MSEE pair. It is a link-layer requirement, not a licence to send 9,000-byte payloads end to end. |
| Jumbo frames (9,000 bytes) | Not supported | Neither on provider-managed circuits nor on ExpressRoute Direct, and neither on private peering nor Microsoft peering. |
Fragmentation Behaviour
- ExpressRoute gateways do not support fragmented packets. A fragmented packet arriving at the gateway is dropped rather than reassembled, so the application must avoid producing IP fragmentation in the first place.
- The supported mitigations are application-side: reduce the sending MTU, rely on Path MTU Discovery (PMTUD), or apply TCP MSS clamping on the on-premises edge router.
- If your workload genuinely requires IP fragmentation support, the documented option is to enable ExpressRoute FastPath, which bypasses the ExpressRoute gateway data plane entirely.
[!CAUTION] The classic MTU failure signature: small requests (a TCP handshake, an ICMP ping, a DNS lookup) succeed, but large transfers — a file copy, a database bulk load, a TLS handshake carrying a large certificate chain — hang or reset. That asymmetry between "small works, large fails" over a hybrid link is the exam's cue for an MTU/fragmentation problem, not a BGP, NSG, or firewall problem.
A multinational financial services institution requires a dedicated hybrid connection to Azure to ingest 80 Gbps of streaming market data into virtual machines. The design mandates line-rate Layer 2 hardware encryption between on-premises edge routers and Microsoft edge devices, zero reliance on third-party carrier exchanges, and storage of cryptographic keys within an Azure Key Vault. Which ExpressRoute connectivity architecture fulfills these specifications?
An enterprise plans to configure ExpressRoute Microsoft Peering to allow on-premises users to access Azure Blob Storage and Microsoft 365 Exchange Online over a dedicated circuit. After configuring the BGP peering session, the on-premises border routers receive no BGP routes from Microsoft, and traffic to Exchange Online fails. What two actions must the network engineer take to establish connectivity?
A network architect observes high latency and CPU throttling on an ErGw3AZ Virtual Network Gateway during peak data replication over an ExpressRoute circuit. The target virtual machines reside in a spoke Virtual Network peered with the hub VNet. Which configuration should the architect implement to achieve line-rate throughput and bypass the gateway data plane for VM traffic?