This section covers the performance and scale architecture of the VMware SD-WAN Edge. It provides recommendations based on tests conducted on the various Edges configured with specific service combinations. It also explains performance and scale data points and how to use them.

Introduction

The tests represent common deployment scenarios to provide recommendations that apply to most deployments. The test data herein are not all-inclusive metrics, nor are they performance or scale limits. There are implementations where the observed performance exceeds the test results and others where specific services, extremely small packet sizes, or other factors can reduce performance below the test results.

Customers are welcome to perform independent tests, and results could vary. However, recommendations based on our test results are adequate for most deployments.

VMware SD-WAN Edge

VMware SD-WAN Edges are zero-touch, enterprise-class appliances that provide secure optimized connectivity to private, public, and hybrid applications as well as compute and virtualized services. VMware SD-WAN Edges perform deep application recognition of traffic flows, performance metrics measurements of underlay transport and apply end-to-end quality of service by applying packet-based link steering and on-demand application remediation, in addition to supporting other virtualized network services.

Throughput Performance Test Topologies

Figure 1. FIGURE 1: Throughput performance test topology for devices 1 Gbps or lower
Figure 2. FIGURE 2: Throughput performance test topology for devices above 1 Gbps

Test Methodology

This subsection details the performance and scale test methodology used to derive the results.

Performance Test Methodology

The testing methodology for Edges uses the industry benchmarking standard RFC 2544 as a framework to execute throughput performance testing. There are specific changes to the type of traffic used and configurations set during testing, described below:

  1. Performance is measured using a fully operational SD-WAN network overlay (DMPO tunnels) test topology in order to exercise the SD-WAN features and obtain results that can be used to appropriately size WAN networks. Testing is conducted using stateful traffic that establishes multiple flows (connections) and are a mix of well-known applications. The number of flows depends on the platform model being tested. Platforms are divided by expected aggregate performance of under 1 Gbps and over 1 Gbps models. Typically, hundreds of flows are needed to fully exercise and determine max throughput of platforms expected to perform under 1 Gbps, and thousands of flows are used to exercise platforms of over 1 Gbps.

    The traffic profiles simulate two network traffic conditions:

    • Large Packet, a 1300-byte condition.
    • IMIX, a mix of packet sizes that average to a 417-byte condition.

    These traffic profiles are used separately to measure maximum throughput per profile.

  2. Performance results are recorded at a packet drop rate (PDR) of 0.01%. The PDR mark provides a more realistic performance result which accounts for normal packet drop that may occur within the SD-WAN packet pipeline in the device. A PDR of 0.01% does not impact application experience even in single link deployment scenarios.
    • The device under test is configured with the following DMPO features; IPsec encrypted using AES-128 and SHA1 for hashing, Application Recognition, link SLA measurements, per-packet forwarding. Business Policy is configured to match all traffic as bulk/low priority to prevent DMPO NACK or FEC from executing and incorrectly altering the traffic generator’s packet count tracking.

Test Results

VMware SD-WAN Edge Performance and Scale Results

Performance metrics are based on the Test Methodology detailed above.

Switched Port Performance: VMware SD-WAN Edges are designed to be deployed as gateway routers between the LAN and the WAN. However, the Edges also provide the flexibility of meeting a variety of other deployment topologies. For example, SD-WAN Edges can have their interfaces configured to operate as switched ports—allowing the switching of LAN traffic between various LAN interfaces without the need for an external device.

An Edge with its interfaces configured as switched ports is ideal for small office deployments where high throughput is not required, as the additional layer of complexity required to handle traffic switching reduces the overall performance of the system. For most deployments, VMware recommends using all routed interfaces.

Note:
  • The Edge device's Maximum Throughput is the sum of throughput across all interfaces of the Edge under test.
  • Overall traffic is the “aggregate” of all traffic flows going to and from an Edge device.
Table 1. Physical Edge Appliances
VMware SD-WAN Edge 510, 510N 510-LTE 520 520V 540 610, 610C, 610N 610-LTE 710-W
Maximum Throughput Large Packet (1300-byte)
Routed Mode All Ports 850 Mbps 850 Mbps 850 Mbps 850 Mbps 1.5 Gbps 850 Mbps 850 Mbps 950 Mbps
Maximum Throughput Internet Traffic (IMIX)
Routed Mode All Ports 300 Mbps 300 Mbps 300 Mbps 300 Mbps 650 Mbps 300 Mbps 300 Mbps 350 Mbps
Routed Mode All Ports with Edge Intelligence activated. 200 Mbps 200 Mbps 200 Mbps 200 Mbps 500 Mbps 200 Mbps 200 Mbps 265 Mbps
Routed Mode All Ports with IPS, Malicious IP Filtering, and Stateful Firewall activated. 150 Mbps 150 Mbps 150 Mbps 150 Mbps 350 Mbps 175 Mbps 175 Mbps 250 Mbps
Routed Mode All Ports with Edge Intelligence, IPS, Malicious IP Filtering, and Stateful Firewall all activated. 150 Mbps 150 Mbps 150 Mbps 150 Mbps 350 Mbps 175 Mbps 175 Mbps 250 Mbps
Other Scale Vectors
Maximum Tunnel Scale 50 50 50 50 100 50 50 50
Flows Per Second 2,400 2,400 2,400 2,400 4,800 2,400 2,400 4,000
Flows Per Second with Edge Intelligence activated 1,200 1,200 1,200 1,200 1,200 1,200 1,200 3,200
Maximum Concurrent Flows 225K 225K 225K 225K 225K 225K 225K 225K
Maximum Concurrent Flows with Edge Intelligence activated. 110K 110K 110K 110K 110K 110K 110K 110K
Maximum Concurrent Flows with IPS, Malicious IP Filtering, and Stateful Firewall activated. 110K 110K 110K 110K 110K 110K 110K 110K
Maximum Concurrent Flows with Edge Intelligence, IPS, Malicious IP Filtering, and Stateful Firewall activated. 110K 110K 110K 110K 110K 110K 110K 110K
Maximum Number of BGP Routes 100K 100K 100K 100K 100K 100K 100K 110K
Maximum Number of Segments 32 32 32 32 32 32 32 32
Maximum Number of NAT Entries 225K 225K 225K 225K 225K 225K 225K 225K
Table 2.
VMware SD-WAN Edge 640, 640C, 640N 680, 680C, 680N 840 2000 3400, 3400C 3800, 3800C 3810
Maximum Throughput Large Packet (1300-byte)
Routed Mode All Ports 5 Gbps 8 Gbps 6 Gbps 15 Gbps 10 Gbps 15 Gbps 15 Gbps
Maximum Throughput Internet Traffic (IMIX)
Routed Mode All Ports 2 Gbps 3 Gbps 2 Gbps 6 Gbps 3.5 Gbps 6.4 Gbps 6.4 Gbps
Routed Mode All Ports with Edge Intelligence activated. 1 Gbps 2 Gbps 1.5 Gbps 5 Gbps 3 Gbps 5 Gbps 5 Gbps
Routed Mode All Ports with IPS and Stateful Firewall activated. 700 Mbps 1.5 Mbps 1 Gbps 3.5 Gbps 1.7 Gbps 3.5 Gbps 3.5 Gbps
Routed Mode All Ports with Edge Intelligence, IPS, and Stateful Firewall all activated. 600 Mbps 1.5 Gbps 800 Mbps 3.5 Gbps 2.5 Gbps 4 Gbps 4 Gbps
Other Scale Vectors
Maximum Tunnel Scale 400 800 400 6,000 4,000 6,000 6,000
Flows Per Second 19,200 19,200 19,200 38,400 38,400 38,400 38,400
Flows Per Second with Edge Intelligence activated 9,600 9,600 9,600 19,200 19,200 19,200 19,200
Maximum Concurrent Flows 1.9M 1.9M 1.9M 3.8M 1.9M 3.8M 3.8M
Maximum Concurrent Flows with Edge Intelligence activated 960K 960K 960K 960K 960K 960K 960K
Maximum Number of Routes 100K 100K 100K 100K 100K 100K 100K
Maximum Number of Segments 128 128 128 128 128 128 128
Maximum Number of NAT Entries 650K 650K 650K 960K 960K 960K 960K
Note:
  • Large Packet performance is based on a large packet (1300-byte) payload with AES-128 encryption and DPI turned on.
  • Internet Traffic (IMIX) performance is based on an average packet size of 417-byte payload with AES-128 encryption and DPI turned on.
  • Edge Intelligence peformance numbers were measured with a 400-byte payload.
  • IPS and Stateful Firewall performance numbers were measured using TREX setup with an average packet size of 400-bytes.
Important: Maximum Tunnel Scale is understood as the total number of tunnels an Edge model can establish at one time with all other sites. However, the maximum number of tunnels an Edge can establish with another Edge or Gateway is 16, regardless of Edge model or type. Each public WAN link an Edge uses establishes a tunnel with each WAN link the peer Edge or Gateway has.

For example: Edge 1 with public WAN links A, B, C, and D connects to Edge 2 with public WAN links E, F, G, and H. Edge 1's WAN link A establishes a tunnel with each of Edge 2's WAN links E, F, G, and H for a total of 4 tunnels for WAN link A to Edge 2. And this follows for Edge 1's other WAN links B, C, and D. Each establishes tunnels with Edge 2's four public WAN links and so four WAN links with 4 tunnels each results in Edge 1 having 16 total tunnels to Edge 2. In this example, no additional tunnels can be established between the two Edges if an additional WAN link is added to either Edge as the maximum has been reached.

Tip: Multiple SD-WAN Edges can be deployed in a cluster for multi-gigabit performance.
Table 3. Edge Maximum Throughput When a Firewall VNF is Actively Service Chained:
Edge Model 520V 620, 620C, 620N 640, 640C, 640N 680, 680C, 680N 840 3400, 3400C 3800, 3800C 3810
Max. Throughput with FW VNF (1300-byte) 100 Mbps 300 Mbps 600 Mbps 1 Gbps 1 Gbps 2 Gbps 3 Gbps 3 Gbps
Table 4. Enhanced High-Availability (HA) Link Performance
Edge Model 510, 510N 510-LTE 520, 520v 540 610, 610C, 610N 610-LTE 710-W
Maximum Throughput (IMIX) Across Enhanced HA Link 220 Mbps 220 Mbps 220 Mbps 480 Mbps 220 Mbps 220 Mbps 260 Mbps
Edge Model 640, 640C, 640N 680, 680C, 680N 840 2000 3400, 3400C 3800, 3800C 3810
Maximum Throughput (IMIX) Across Enhanced HA Link 1 Gbps 2 Gbps 1 Gbps 4 Gbps 2.5 Gbps 5 Gbps 5 Gbps
Important: Performance with Edge Intelligence activated:
  • There is a performance impact of up to 20% when analytics are activated.
  • Flow capacity is reduced by half when analytics are activated due to the additional memory and processing required for analysis.

Platform Independent Edge Scale Numbers

The Edge Scale numbers listed in the following table are platform independent and are valid for all Edge models, both hardware and virtual.
Note: The listed maximum value for each feature represents the supported limits that have been tested and verified by VMware. In some cases, customers may exceed values higher than that is listed in the table. If a customer exceeds the published maximum value, the environment may work, but VMware cannot guarantee that it would.
Feature Supported Number
IPv4 IPv6
Maximum number of Port Forwarding rules on a single segment 128 128
Maximum number of Port Forwarding rules across 16 segments 128 128
Maximum number of Port Forwarding rules across 128 segments 128 128
Maximum number of Outbound Firewall Rules on a single segment 2040 2040
Maximum number of Outbound Firewall Rules across 16 segments 2040 2040
Maximum number of Outbound Firewall Rules across 128 segments 2040 2040
Maximum number of 1:1 NAT rules on a single segment 128 128
Maximum number of 1:1 NAT rules across 16 segments 128 128
Maximum number of 1:1 NAT rules across 128 segments 128 128
Maximum number of LAN side NAT rules on a single segment 256 -
Maximum number of LAN side NAT rules across 16 segments 256 -
Maximum number of LAN side NAT rules across 128 segments 256 -
Maximum number of Object Groups (1000 business policies, each business policy assigned to one object group, each object group supports 255 address groups)  1000 1000

Virtual Edge

Table 5. Private Cloud (Hypervisors)
Edge Device Maximum Throughput Maximum Number of Tunnels Flows Per Second Maximum Concurrent Flows Maximum Number of Routes Maximum Number of Segments
ESXi Virtual Edge (2-core, VMXNET3)

1.5 Gbps (1300-byte)

900 Mbps (IMIX)

50 2400 240K 35K 128
KVM Virtual Edge (2-core, Linux Bridge)

800 Mbps (1300-byte)

250 Mbps (IMIX)

50 2400 240K 35K 128
KVM Virtual Edge (2-core, SR-IOV)

1.5 Gbps (1300-byte)

900 Mbps (IMIX)

50 2400 240K 35K 128
ESXi Virtual Edge (4-core, VMXNET3)

4 Gbps (1300-byte)

1.5 Gbps (IMIX)

400 4800 480K 35K 128
ESXi Virtual Edge (4-core, SR-IOV)

5 Gbps (1300-byte)

1.5 Gbps (IMIX)

400 4800 480K 35K 128
KVM Virtual Edge (4-core, Linux Bridge)

1 Gbps (1300-byte)

350 Mbps (IMIX)

400 4800 480K 35K 128
KVM Virtual Edge (4-core, SR-IOV)

4 Gbps (1300-byte)

1.5 Gbps (IMIX)

400 4800 480K 35K 128
ESXi Virtual Edge (8-core, VMXNET3)

6 Gbps (1300-byte)

2 Gbps (IMIX)

800 28800 1.9M 35K 128
ESXi Virtual Edge (8-core, SR-IOV)

6 Gbps (1300-byte)

3 Gbps (IMIX)

800 28800 1.9M 35K 128
KVM Virtual Edge (8-core, SR-IOV

6.5 Gbps (1300-byte)

3.2 Gbps (IMIX)

800 28800 1.9M 35K 128
2 vCPU 4vCPU 8vCPU 10vCPU
Minimum Memory (DRAM) 8 GB 16 GB 32 GB 32 GB
Minimum Storage 8 GB 8 GB 16 GB 16 GB
Supported Hypervisors

Software version 4.0 and above:

  • ESXi 6.5U1, 6.7U1, 7.0
  • KVM Ubuntu 16.04 and 18.04
Supported Public Cloud AWS, Azure, GCP, and Alibaba
Support Network I/O SR-IOV, VirtIO, VMXNET3
Recommended Host Settings

CPUs at 2.0 GHz or higher

CPU configuration:

  • AES-NI activated.
  • Power savings deactivated
  • CPU turbo activated
  • Hyper-threading deactivated
  • Minimum instructions sets: SSE3, SSE4, and RDTSC.
  • Recommended instruction sets: AVX2 or AVX512
VMware ESXi required settings:
  • CPU reservation: Maximum
  • CPU shares: High
  • Memory reservation: Maximum
  • Latency sensitivity: High
Note: Performance metrics are based on a system using an Intel ® Xeon ® CPU E5-2683 v4 at 2.10 GHz (AES-NI).

Public Cloud

Table 6. Amazon Web Services (AWS)
AWS Instance Type c5.large c5.xlarge c5.2xlarge c5.4xlarge
Maximum Throughput

100 Mbps (1300-byte)

50 Mbps (IMIX)

200 Mbps (1300-byte)

100 Mbps (IMIX)

1.5 Gbps (1300-byte)

450 Mbps (IMIX)

4 Gbps (1300-byte)

1 Gbps (IMIX)

Maximum Tunnels 50 400 800 2,000
Flows Per Second 1,200 2,400 4,800 9,600
Maximum Concurrent Flows 125,000 250,000 550,000 1.9M
Maximum Number of Routes 35,000 35,000 35,000 35,000
Maximum Number of Segments 128 128 128 128
Note: c5.2xlarge and c5.4xlarge performance and scale numbers are based on AWS Enhanced Networking (ENA SR-IOV drivers) being ‘activated’.
Table 7. Microsoft Azure (Without Accelerated Networking)
Azure VM Series D2d v4 D4d v4 D8d v4 D16d v4
Maximum Throughput

100 Mbps (1300-byte)

50 Mbps (IMIX)

200 Mbps (1300-byte)

100 Mbps (IMIX)

1 Gbps (1300-byte)

450 Mbps (IMIX)

1 Gbps (1300-byte)

450 Mbps (IMIX)

Maximum Tunnels 50 400 800 2000
Flows Per Second 1,200 2,400 4,800 4,800
Maximum Concurrent Flows 125,000 250,000 550,000 550,000
Maximum Number of Routes 35,000 35,000 35,000 35,000
Maximum Number of Segments 128 128 128 128
Table 8. Microsoft Azure (With Accelerated Networking)
Azure VM Series Ds3 v2 Ds4 v2 Ds5 v2 D4d v5 D8d v5 D16d v5
Maximum Throughput

2.5 Gbps (1300-byte)

1.5 Gbps (IMIX)

5.3 Gbps (1300-byte)

2.7 Gbps (IMIX)

6.5 Gbps (1300-byte)

3.1 Gbps (IMIX)

4.5 Gbps (1300-byte)

1.3 Gbps (IMIX)

6.3 Gbps (1300-byte)

2.7 Gbps (IMIX)

6.4 Gbps (1300-byte)

2.9 Gbps (IMIX)
Maximum Tunnels 400 800 2000 400 800 2000
Flows Per Second 2,400 4,800 4,800 2,400 4,800 4,800
Maximum Concurrent Flows 250,000 550,000 550,000 250,000 550,000 550,000
Maximum Number of Routes 35,000 35,000 35,000 35,000 35,000 35,000
Maximum Number of Segments 128 128 128 128 128 128
Note:
  • Azure Accelerated Networking is supported only from release 5.4.0.
  • Accelerated Networking is supported only on Connect-X4 and Connect-X5 NICs.
Table 9. Google Cloud Platform
GCP Instance Type n2-highcpu-4 n2-highcpu-8 n2-highcpu-16
Maximum Throughput

850 Mbps (1300-byte)

500 Mbps (IMIX)

4.5 Gbps (1300-byte)

1.6 Gbps (IMIX)

6.5 Gbps (1300-byte)

1.9 Gbps (IMIX)

Maximum Tunnels 50 400 800
Flows Per Second 1,200 2,400 4,800
Maximum Concurrent Flows 125,000 250,000 550,000
Maximum Number of Routes 35,000 35,000 35,000
Maximum Number of Segments 128 128 128

Use of DPDK on VMware SD-WAN Edges

To improve packet throughput performance, VMware SD-WAN Edges take advantage of Data Plane Development Kit (DPDK) technology. DPDK is a set of data plane libraries and drivers provided by Intel for offloading TCP packet processing from the operating system kernel to processes running in user space and results in higher packet throughput. For more details, see https://www.dpdk.org/.

Edge hardware models 620 and higher and all virtual Edges use DPDK by default on their routed interfaces. Edges do not use DPDK on their switched interfaces. A user cannot activate or deactivate DPDK for an Edge interface.