To run Anbox Cloud you need to fullfil a few minimum requirements which are a bit different depending on which kind of deployment you choose.
The following two different deployment variants and their minimum requirements will be covered below:
- The Anbox Cloud Appliance
- Juju based deployments
See the overview for a explanation of the differences between both variants.
The following requirements apply to all variants of Anbox Cloud
Ubuntu Advantage Token
After registering to Anbox Cloud, you should have received an Ubuntu Advantage for Applications token. If you haven’t received one, please contact support or your Canonical account representative as you’ll need it to deploy Anbox Cloud.
Anbox Cloud is only supported on the Ubuntu variant of the Linux kernel starting with version 5.4. The GA kernel for Ubuntu 18.04 (based on 4.15) is not supported. The following table gives an overview of the available kernel versions for the different supported clouds:
|CLOUD||KERNEL VERSION||UBUNTU PACKAGES|
|AWS||>= 5.4||linux-aws, linux-modules-extra-aws|
|GCE||>= 5.4||linux-gcp, linux-modules-extra-gcp|
|Azure||>= 5.4||linux-azure, linux-modules-extra-azure|
|OCI||>= 5.4||linux-oracle, linux-modules-extra-oracle|
Anbox Cloud Appliance
The Anbox Cloud Appliance has the following minimum hardware requirements:
- 64 bit x86 or Arm CPU with >= 4 CPU cores
- 8 GB of memory
- 40 GB of disk space for the OS
- optional, but strongly recommend: >= 50GB block volume to host container storage
The above defines a minimum of what is necessary to run the Anbox Cloud Appliance. As Anbox Cloud is dependent on available resources to launch it’s Android containers the available resources dictate the maximum number of possible containers. See Capacity Planing for an explanation on how to plan for a specific capacity on your appliance.
On public clouds it’s recommend to always allocate an additional storage volume for the container storage. If no additional storage volume is available, the appliance will create an on disk image it will use for the container storage. This is sufficient for very simple cases but does not provide optimal performance and will slow down operations and container startup time.
For external access to the Anbox Cloud Appliance a couple of network ports needs to be expose on the machine it’s running on. The following table lists all ports.
|80||tcp||mandatory||HTTP (redirects to HTTPS on port 443)|
|60000-60100||udp||mandatory||TURN relay ports|
|10000-11000||udp||optional||Service endpoints exposed by AMS|
|10000-11000||tcp||optional||Service endpoints exposed by AMS|
How to allow incoming traffic on the listed ports is different depending on the cloud used. Please consult the documentation of the cloud for further information on how to change the firewall.
Juju based deployments
Anbox Cloud deployments are managed by Juju. They can be crwated on all the supported clouds as well as manually provided machines as long as they follow the required minimums.
While you can run Anbox Cloud on a single machine, we strongly recommend the following setup for a production environment:
|0||amd64||4||4GB||50GB SSD||no||Hosts the Juju controller|
|1||amd64||4||8GB||100GB SSD||no||Host the management layer of Anbox Cloud|
|2||amd64 or arm64||8||16GB||200GB NVMe||optional||LXD worker node. Hosts the actual Anbox containers|
The specified number of cores and RAM is only the minimum required to run Anbox Cloud at a sensible performance.
More CPU cores and more RAM on the machine hosting LXD will allow to run a higher number of containers. See Capacity Planing for an introduction of how many resources are necessary to host a specific number of containers.
Applications not maintained by Anbox Cloud may have different hardware recommendations:
- etcd: https://etcd.io/docs/v3.4.0/op-guide/hardware/
- HAProxy (load balancer for the Stream Gateway and the dashboard): https://www.haproxy.com/documentation/hapee/latest/installation/getting-started/os-hardware/#hardware-requirements
Please note that these are just baselines and should be adapted to your worload. No matter the application, measuring performances is always important.
Last updated 22 hours ago.