Hardware requirements
While Warren supports a majority of commodity hardware there are still minimum requirements in place that will provide a redundant and reliable IaaS cluster.
Hardware specifications below are a good indication of a production grade setup to start with and to grow upon.
Recommended hardware
The following hardware recommendation is optimal for most commercial use cases and launching a regional cloud business.
| Type | Nodes | Configuration | |
|---|---|---|---|
| Control |  |
3 x Warren Control Nodes
for resilient IaaS solution |
|
| Compute | |
3 x Compute nodes
used to run customer workloads (minimum of 3 nodes gives you live migrations in case of hardware failure) |
|
| Storage | |
3 x Ceph nodes
for distributed block storage immune to most hardware failures |
|
| Object Storage (S3) (optional) |
 |
2 x Ceph nodes (in addition to standard general storage)
for distributed object storage immune to most hardware failures |
|
| Network |  |
2 x TOR & 1 x management switch
for redundant high speed network |
|
- Recommended reading on how to be smart about choosing the right hardware: Aligning Hardware to Your IaaS Business Plan
- More information about Warren cloud platform control plane, network topology and technologies used could be found here: https://warrenio.atlassian.net/wiki/spaces/WARP/pages/300875801/Warren+Hardware+Recommendation
Why Warren needs 3 physical control nodes to operate?
Since the software is built exclusively for production grade setups, Warren team has decided to deliver only highly redundant and resilient deployments especially if the initial purpose of the setup is a pilot or proof of concept.
What actually runs on the 3 physical control nodes?
Delivering a turnkey IaaS solution involves much more than for instance vanilla OpenStack or CloudStack provides. The following breakdown should give an high level overview of some of the components running on the 3 Warren Control Nodes.
NixOS operating system is specifically designed to declaratively deploy identical cloud systems for maximal fault tolerance. Every time a new release is deployed the prior version will be discarded and every cloud provider will have the fresh features introduced with zero downtime (usually twice every month).
Some of the open-source software hosted with the control plane
- Tungsten Fabric – Software Defined Networking.
- Ceph – Software Defined Storage.
- Prometheus, Loki & Grafana – Monitoring managed workloads remotely and locally.
- Apache Mesos & Marathon – Compute cluster manager that handles workloads in a distributed environment through dynamic resource sharing and isolation.
- Kong – Gateway for presenting endpoints of most Warren components from a single root API.
- Consul – Service discovery & key-value store
Warren platform components / microservices
- Base operator – Virtual machine and compute instance management.
- User Interface – Each UI user gets an individual UI Docker container of JupyterLab for isolation.
- User Resource Interface – User management, token management and managed services management.
- Storage Manager – Blocks storage management, object storage management.
- Charging Engine – Collecting user resource usage (currently on hour precision)
- Network Manager – VPC, FIP and Load Balancer management.
- Payment – Invoices, payment gateways, pricing management.
Rack planning
 |
Configuration |
BASE
12 nodes
STANDALONE RACK |
GROWTH
32 nodes
STANDALONE RACK |
SCALE
31 nodes
3 RACK MINIMUM |
|---|---|---|---|---|
| Aggregation node |  |
|||
| Control nodes |  |
|
 |
|
| Compute nodes |  |
 |
|
|
| Network nodes |  |
|
|
|
| Block storage nodes | |
 |
|
|
| Object storage nodes (optional) |
 |
|
||
| * Every other rack |  |
|||
Routers are not included.
Juniper routers are required for IPv6 support.
