post

Setting up a 100GbE PVRDMA Network on vCenter 7

After writing my last article on Getting NVIDIA NGC containers to work with VMware PVRDMA networks I had a couple of people ask me “How do I set up PVRDMA networking on vCenter?” These are the steps that I took to set up PVRDMA networking in my lab.

RDMA over Converged Ethernet (RoCE) is a network protocol that allows remote direct memory access (RDMA) over an Ethernet network. It works by encapsulating an Infiniband (IB) transport packet and sending it over Ethernet. If you’re working with network applications that require high bandwidth and low latency, RDMA will give you lower latency, higher bandwidth, and a lower CPU load than an API such as Berkeley sockets.

Full disclosure: I used to work for a startup called Bitfusion, and that startup was bought by VMware, so I now work for VMware. At Bitfusion we developed a technology for accessing hardware accelerators, such as NVIDIA GPUs, remotely across networks using TCP/IP, Infiniband, and PVRDMA. I still work on the Bitfusion product at VMware, and spend a lot of my time getting AI and ML workloads to work across networks on virtualized GPUs.

In my lab I’m using Mellanox Connect/X5 and ConnectX/6 cards on hosts that are running ESXi 7.0.2 and vCenter 7.0.2. The cards are connected to a Mellanox Onyx MSN2700 100GbE switch.

Since I’m working with Ubuntu 18.04 and 20.04 virtual machines (VMs) in a vCenter environment, I have a couple of options for high-speed networking:

  • I can use PCI passthrough to pass the PCI network card directly through to the VM and use the network card’s native drivers on the VM to set up a networking stack. However this means that my network card is only available to a single VM on the host, and can’t be shared between VMs. It also breaks vMotion (the ability to live-migrate the VM to another host) since the VM is tied to a specific piece of hardware on a specific host. I’ve set this up in my lab but stopped doing this because of the lack of flexibility and because we couldn’t identify any performance difference compared to SR-IOV networking.
  • I can use SR-IOV and Network Virtual Functions (NVFs) to make the single card appear as if it’s multiple network cards with multiple PCI addresses, pass those through to the VM, and use the network card’s native drivers on the VM to set up a networking stack. I’ve set this up in my lab as well. I can share a single card between multiple VMs and the performance is similar to PCI passthough. The disadvantages are that setting up SR-IOV and configuring the NVFs is specific to a card’s model and manufacturer, so what works in my lab might not work in someone else’s environment.
  • I can set up PVRDMA networking and use the PVRDMA driver that comes with Ubuntu. This is what I’m going to show how to do in this article.

Set up your physical switch

First, make sure that your switch is set up correctly. On my Mellanox Onyx MSN2700 100GbE switch that means:

  • Enable the ports you’re connecting to.
  • Set the speed of each port to 100G.
  • Set auto-negotiation for each link.
  • MTU: 9000
  • Flowcontrol Mode: Global
  • LAG/MLAG: No
  • LAG Mode: On

Set up your virtual switch

vCenter supports Paravirtual RDMA (PVRDMA) networking using Distributed Virtual Switches (DVS). This means you’re setting up a virtual switch in vCenter and you’ll connect your VMs to this virtual switch.

In vCenter navigate to Hosts and Clusters, then click the DataCenter icon (looks like a sphere or globe with a line under it). Find the cluster you want to add the virtual switch to, right click on the cluster and select Distributed Switch > New Distributed Switch.

  • Name: “rdma-dvs”
  • Version: 7.0.2 – ESXi 7.0.2 and later
  • Number of uplinks: 4
  • Network I/O control: Disabled
  • Default port group: Create
  • Port Group Name: “VM 100GbE Network”

Figure out which NIC is the right NIC

  • Go to Hosts and Clusters
  • Select the host
  • Click the Configure tab, then Networking > Physical adapters
  • Note which NIC is the 100GbE NIC for each host

Add Hosts to the Distributed Virtual Switch

  • Go to Hosts and Clusters
  • Click the DataCenter icon
  • Select the Networks top tab and the Distributed Switches sub-tab
  • Right click “rdma-dvs”
  • Click “Add and Manage Hosts”
  • Select “Add Hosts”
  • Select the hosts. Use “auto” for uplinks.
  • Select the physical adapters based on the list you created in the previous step, or find the Mellanox card in the list and add it. If more than one is listed, look for the card that’s “connected”.
  • Manage VMkernel adapters (accept defaults)
  • Migrate virtual machine networking (none)

Tag a vmknic for PVRDMA

  • Select an ESXi host and go to the Configure tab
  • Go to System > Advanced System Settings
  • Click Edit
  • Filter on “PVRDMA”
  • Set Net.PVRDMAVmknic = "vmk0"

Repeat for each ESXi host.

Set up the firewall for PVRDMA

  • Select an ESXi host and go to the Configure tab
  • Go to System > Firewall
  • Click Edit
  • Scroll down to find pvrdma and check the box to allow PVRDMA traffic through the firewall.

Repeat for each ESXi host.

Set up Jumbo Frames for PVRDMA

To enable jumbo frames a vCenter cluster using virtual switches you have to set MTU 9000 on the Distributed Virtual Switch.

  • Click the Data Center icon.
  • Click the Distributed Virtual Switch that you want to set up, “rdma-dvs” in this example.
  • Go to the Configure tab.
  • Select Settings > Properties.
  • Look at Properties > Advanced > MTU. This should be set to 9000. If it’s not, click Edit.
  • Click Advanced.
  • Set MTU to 9000.
  • Click OK.

Add a PVRDMA NIC to a VM

  • Edit the VM settings
  • Add a new device
  • Select “Network Adapter”
  • Pick “VM 100GbE Network” for the network.
  • Connect at Power On (checked)
  • Adapter type PVRDMA (very important!)
  • Device Protocol: RoCE v2

Configure the VM

For Ubuntu:

sudo apt-get install rdma-core infiniband-diags ibverbs-utils

Tweak the module load order

In order for RDMA to work the vmw_pvrdma module has to be loaded after several other modules. Maybe someone else knows a better way to do this, but the method that I got to work was adding a script /usr/local/sbin/rdma-modules.sh to ensure that Infiniband modules are loaded on boot, then calling that from /etc/rc.local so it gets executed at boot time.

#!/bin/bash
# rdma-modules.sh
# modules that need to be loaded for PVRDMA to work
/sbin/modprobe mlx4_ib
/sbin/modprobe ib_umad
/sbin/modprobe rdma_cm
/sbin/modprobe rdma_ucm

# Once those are loaded, reload the vmw_pvrdma module
/sbin/modprobe -r vmw_pvrdma
/sbin/modprobe vmw_pvrdma

Once that’s done just set up the PVRDMA network interface the same as any other network interface.

Testing the network

To verify that I’m getting something close to 100Gbps on the network I use the perftest package.

To test bandwith I pick two VMs on different hosts. On one VM I run:

$ ib_send_bw --report_gbits

On the other VM I run the same command plus I add the IP address of the PVRDMA interface on the first machine:

$ ib_send_bw --report_gbits 192.168.128.39

That sends a bunch of data across the network and reports back:

So I’m getting an average of 96.31Gbps over the network connection.

I can also check the latency using the ib_send_lat:

Hope you find this useful.

post

Getting NVIDIA NGC containers to work with VMware PVRDMA networks

NVIDIA publishes a set of NVIDIA GPU-accelerated Containers (NGC) with applications and frameworks for machine learning, deep learning, and high-performance computing.

VMware developed a platform that allows people and companies to create their own private clouds. For customers with high-speed, low-latency networking requirements they offer a couple of different networking options, one of which is PVRDMA (ParaVirtualized Remote Direct Memory Access) networking.

Full disclosure: I used to work for a startup called Bitfusion, and that startup was bought by VMware, so I now work for VMware. At Bitfusion we developed a technology for accessing hardware accelerators, such as NVIDIA GPUs, remotely across networks using TCP/IP, Infiniband, and PVRDMA. I still work on the Bitfusion product at VMware, and spend a lot of my time getting AI and ML workloads to work across networks on virtualized GPUs.

OpenFabrics Enterprise Distribution (OFED) is open-source software for RDMA applications which includes a set of drivers for high-speed network cards to enable RDMA/Infiniband networking. Some NVIDIA NGC containers ship with Mellanox OFED (MOFED) installed. NVIDIA bought Mellanox in 2020, and MOFED is NVIDIA’s distribution of OFED with all of the non-Mellanox drivers removed. OFED includes support for PVRDMA, but MOFED does not.

NVIDIA containers are based on Ubuntu base images. Ubuntu ships its own RDMA drivers in a package called rdma-core. The Ubuntu rdma-core package contains the open source drivers and utilities needed to work with VMware PVRDMA networking.

The Ubuntu rdma-core package contains the open source drivers and utilities needed to work with VMware PVRDMA networking.

Ideally you should only install the RDMA network package that you need, either MOFED or OFED or rdma-core, but not more than one of them. In fact, if you try installing more than one you will have problems. Therefore, if you’re going to use NGC containers on a PVRDMA network you should first remove the MOFED packages and then add the rdma-core packages.

Luckily you can start an NGC container and see if MOFED is installed or not and see what version is installed. If I start the NGC container for Tensor RT:

docker run -it --rm -u root nvcr.io/nvidia/tensorrt:19.09-py3

I can see that it’s based on Ubuntu 18.04 “bionic”:

root@2e70d41e1187:/workspace# cat /etc/os-release
NAME="Ubuntu"
VERSION="18.04.3 LTS (Bionic Beaver)"
ID=ubuntu
ID_LIKE=debian
PRETTY_NAME="Ubuntu 18.04.3 LTS"
VERSION_ID="18.04"
HOME_URL="https://www.ubuntu.com/"
SUPPORT_URL="https://help.ubuntu.com/"
BUG_REPORT_URL="https://bugs.launchpad.net/ubuntu/"
PRIVACY_POLICY_URL="https://www.ubuntu.com/legal/terms-and-policies/privacy-policy"
VERSION_CODENAME=bionic
UBUNTU_CODENAME=bionic

If I look inside /opt/mellanox/DEBS/ I can see if any MOFED .deb files are installed:

root@2e70d41e1187:/workspace# ls -al /opt/mellanox/DEBS/
total 64
drwxrwxr-x 15 root root 4096 Aug 27  2019 .
drwxr-xr-x  3 root root 4096 Sep 13  2019 ..
drwxrwxr-x  2 root root 4096 Aug 27  2019 3.4-1.0.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 3.4-2.0.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.0-1.0.1
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.0-2.0.0
lrwxrwxrwx  1 root root    9 Aug 27  2019 4.0-2.0.2 -> 4.0-2.0.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.1-1.0.2
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.2-1.0.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.2-1.2.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.3-1.0.1
lrwxrwxrwx  1 root root    9 Aug 27  2019 4.3-3.0.2 -> 4.3-1.0.1
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.4-1.0.0
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.4-2.0.7
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.5-1.0.1
drwxrwxr-x  2 root root 4096 Aug 27  2019 4.6-1.0.1
lrwxrwxrwx  1 root root    9 Aug 27  2019 5.0-0 -> 5.0-1.1.8
drwxrwxr-x  2 root root 4096 Aug 27  2019 5.0-1.1.8
-rwxrwxr-x  1 root root  546 Aug 27  2019 add_mofed_version.sh

In this case there are Mellanox MOFED packages installed. If I look inside these directories (ls -1 /opt/mellanox/DEBS/*) I can see that the packages installed from MOFED are:

  • ibverbs-utils
  • libibverbs-dev
  • libibverbs1
  • libmlx5-1

These are MOFED versions of packages installed in this specific container. A different NGC container might contain these MOFED packages, or different MOFED packages, or no MOFED packages at all.

There are versions of these same packages in Ubuntu repos, and the Ubuntu versions conflict with the MOFED versions. To use the Ubuntu versions, first remove the MOFED packages:

root@2e70d41e1187:/workspace# apt-get purge -y ibverbs-utils libibverbs-dev libibverbs1 libmlx5-1
Reading package lists... Done
Building dependency tree
Reading state information... Done
The following packages will be REMOVED:
  ibverbs-utils* libibverbs-dev* libibverbs1* libmlx5-1*
0 upgraded, 0 newly installed, 4 to remove and 23 not upgraded.
After this operation, 1523 kB disk space will be freed.
(Reading database ... 18622 files and directories currently installed.)
Removing ibverbs-utils (41mlnx1-OFED.4.4.1.0.0.44100) ...
Removing libibverbs-dev (41mlnx1-OFED.4.4.1.0.0.44100) ...
Removing libmlx5-1 (41mlnx1-OFED.4.4.0.1.7.44100) ...
Removing libibverbs1 (41mlnx1-OFED.4.4.1.0.0.44100) ...
Processing triggers for libc-bin (2.27-3ubuntu1) ...
(Reading database ... 18449 files and directories currently installed.)
Purging configuration files for libmlx5-1 (41mlnx1-OFED.4.4.0.1.7.44100) ...

You can see in the output above that the packages that I removed have the name “OFED” in them, indicating that they came from MOFED/OFED, not Ubuntu. If I reinstall using rdma-core and the other packages I need:

apt-get update && apt-get install -y --reinstall \
    -t bionic rdma-core libibverbs1 ibverbs-providers \
    infiniband-diags ibverbs-utils libcapstone3

This installs everything from the Ubuntu repositories for the “bionic” version, which is the version of Ubuntu that this NGC container is based on. (Which we determined back in step 1.)

The -t flag is necessary because I’ve found that some NGC containers mix code from the repositories of different versions of Ubuntu, and we only want to install packages from the base Ubuntu version, which is “bionic” in this particular case.

At this point the container is ready to use PVRDMA connections.

However, I also want to connect to a remote Bitfusion server across a PVRDMA network and use a pool of GPUs for my TensorRT work, so I also install the Bitfusion client:

wget https://packages.vmware.com/bitfusion/ubuntu/18.04/bitfusion-client-ubuntu1804_3.0.0-11_amd64.deb

apt-get install -y ./bitfusion-client-ubuntu1804_3.0.0-11_amd64.deb

To create a new container with all of these changes I just have to whip up a small Dockerfile:

# Base this container on the NGC container you want to use
FROM nvcr.io/nvidia/tensorrt:19.09-py3

# Remove the MOFED packages that are installed,
# determined by running “ls -1 /opt/mellanox/DEBS/*”
RUN apt-get purge -y ibverbs-utils libibverbs-dev \
    libibverbs1 libmlx5-1

# Install the Ubuntu RDMA packages using the
# UBUNTU_CODENAME from /etc/os-release
# as the -t argument.
RUN apt-get update && apt-get install -y --reinstall \
    -t bionic \
    rdma-core libibverbs1 ibverbs-providers \
    infiniband-diags ibverbs-utils libcapstone3

# Install the Bitfusion 3.0.0 client software for Ubuntu 18.04
RUN wget https://packages.vmware.com/bitfusion/ubuntu/18.04/bitfusion-client-ubuntu1804_3.0.0-11_amd64.deb

RUN apt-get install -y ./bitfusion-client-ubuntu1804_3.0.0-11_amd64.deb

To build an image using this Dockerfile:

mkdir -p ~/build
docker build -t tensorrt:19.09-py3-pvrdma -f Dockerfile ~/build

Run this image:

docker run -it --rm -u root --network host \
    tensorrt:19.09-py3-pvrdma

In this instance I’m passing the host’s network through to the container. Assuming that the host already has PVRDMA networking set up correctly, I can use that PVRDMA network inside the NGC container. With the Bitfusion client in the container I can run TensorRT and access GPUs from a remote pool of GPUs across a PVRDMA network.

Hope you find this useful.