NVIDIA BlueField DPU BSP v4.2.2
1.0

Multi-Host

Warning

This is only applicable to DPUs running on multi-host model.

In multi-host mode, each host interface can be divided into up to 4 independent PCIe interfaces. All interfaces would share the same physical port, and are managed by the same multi-physical function switch (MPFS). Each host would have its own e-switch and would control its own traffic.

Multi-host.png

Similar to Kernel Representors Model, each host here has an uplink representor, PF representor, and VF representors (if SR-IOV is enabled). There are 8 sets of representors (uplink/PF; see example code). For each host to work with OVS offload, the corresponding representors must be added to the OVS bridge.

Copy
Copied!
            

139: p0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master ovs-system state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b2 brd ff:ff:ff:ff:ff:ff 140: p1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b3 brd ff:ff:ff:ff:ff:ff 141: p2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master ovs-system state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b4 brd ff:ff:ff:ff:ff:ff 142: p3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b5 brd ff:ff:ff:ff:ff:ff 143: p4: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b6 brd ff:ff:ff:ff:ff:ff 144: p5: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b7 brd ff:ff:ff:ff:ff:ff 145: p6: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b8 brd ff:ff:ff:ff:ff:ff 146: p7: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:b9 brd ff:ff:ff:ff:ff:ff 147: pf0hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq master ovs-system state UP group default qlen 1000 link/ether 86:c5:8a:b7:7c:84 brd ff:ff:ff:ff:ff:ff 148: pf1hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 6e:ea:1b:84:88:49 brd ff:ff:ff:ff:ff:ff 149: pf2hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 92:ec:99:cb:d7:23 brd ff:ff:ff:ff:ff:ff 150: pf3hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0e:0d:8e:03:2e:27 brd ff:ff:ff:ff:ff:ff 151: pf4hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 5e:42:af:05:67:93 brd ff:ff:ff:ff:ff:ff 152: pf5hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 96:e4:69:4c:b7:7f brd ff:ff:ff:ff:ff:ff 153: pf6hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 5e:67:33:c0:35:05 brd ff:ff:ff:ff:ff:ff 154: pf7hpf: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 12:29:7d:56:07:3e brd ff:ff:ff:ff:ff:ff

The following is an example of adding all representors to OVS:

Copy
Copied!
            

Bridge armBr-3 Port armBr-3 Interface armBr-3 type: internal Port p3 Interface p3 Port pf3hpf Interface pf3hpf Bridge armBr-2 Port p2 Interface p2 Port pf2hpf Interface pf2hpf Port armBr-2 Interface armBr-2 type: internal Bridge armBr-5 Port p5 Interface p5 Port pf5hpf Interface pf5hpf Port armBr-5 Interface armBr-5 type: internal Bridge armBr-7 Port pf7hpf Interface pf7hpf Port armBr-7 Interface armBr-7 type: internal Port p7 Interface p7 Bridge armBr-0 Port p0 Interface p0 Port armBr-0 Interface armBr-0 type: internal Port pf0hpf Interface pf0hpf Bridge armBr-4 Port p4 Interface p4 Port pf4hpf Interface pf4hpf Port armBr-4 Interface armBr-4 type: internal Bridge armBr-1 Port armBr-1 Interface armBr-1 type: internal Port p1 Interface p1 Port pf1hpf Interface pf1hpf Bridge armBr-6 Port armBr-6 Interface armBr-6 type: internal Port p6 Interface p6 Port pf6hpf Interface pf6hpf ovs_version: "2.13.1"

For now, users can get the representor-to-host PF mapping by comparing the MAC address queried from host control on the Arm-side and PF MAC on the host-side. In the following example, the user knows p0 is the uplink representor for p6p1 as the MAC address is the same.

From Arm:

Copy
Copied!
            

# cat /sys/class/net/p0/smart_nic/pf/config MAC : 0c:42:a1:70:1d:9a MaxTxRate : 0 State : Up

From host:

Copy
Copied!
            

# ip addr show p6p1 3: p6p1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc mq state UP group default qlen 1000 link/ether 0c:42:a1:70:1d:9a brd ff:ff:ff:ff:ff:ff

The implicit mapping is as follows:

  • PF0, PF1 = host controller 1

  • PF2, PF3 = host controller 2

  • PF4, PF5 = host controller 3

  • PF6, PF7 = host controller 4

Warning

The maximum SF or VF count across all hosts is limited to 488 in total. The user can divide 488 VFs/SFs to single or multiple controllers as desired.

© Copyright 2023, NVIDIA. Last updated on Nov 21, 2023.