The bare metal cloud uses a flat network architecture, with almost no aggregation layer and is optimized for east-west traffic as well as north-south. Each switch on the path between instances is guaranteed to sustain line rate performance on all links at all times.
- The bare metal network is completely dynamic – any two points in the network can be connected at any time.
- It uses cut-through switching to achieve wire speed latency.
- It can create unlimited layer 2 broadcast domains for each user.
- it automatically assigns IPs and seamlessly manages instance migrations.
- It can be securely connected to existing on-premises networks or VPNs.
- Instance ports can be aggregated to provide up to 40 Gbps to a single destination.
All Bare Metal, All Software Defined
When an API call is made or an action takes place in the Control Center, the bare metal network automatically creates a secure tunnel between the two physical ports, linking them via an isolated layer 2 pseudo-wire connection. This tunnel is a bare metal environment in which VLANS can be driven and traffic can broadcast. When a new port is added to the same LAN, the layer 2 network is expanded, just like a physical on-premises network. The Bare Metal Network provides lower latency, more elasticity and improved security compared to traditional cloud networks. It runs entirely on bare metal - there is no virtual or other types of software switching involved.
Power User Design Capabilities
Designing a complex network architecture in the Control Center is as simple as drag & drop. The interface allows building networks as easy as drawing a wireframe on paper. It literally takes minutes to:
Hours Saved on Management
The bare metal network automates repetitive tasks and makes it easy for DevOps, sysadmins and network engineers to build elegantly secure and simple systems:
- IPs are allocated or migrated automatically when infrastructures are built or scaled.
- Networks have customizable labels and colors for easy identification.
- Every action performed in the Control Center can be automated via the API.
The World's Fastest Cloud Network on Bare Metal
The bare metal network delivers lower latency and higher throughput than any other
public cloud network as it was designed from the ground up for performance:
Virtualized public clouds use software switching, which takes milliseconds to analyze each packet. The Bare Metal Network uses exclusively bare metal nanosecond switching. Packets are analyzed and sent on their way within nanoseconds, making it tens of times faster than any other cloud network.
Unlike store-and-forward, cut-through switching only analyzes the first few bytes of a packet in an ASIC before forwarding it to the appropriate port. It doesn’t introduce any latency, so the packets are always transferred through the network at wire speed.
High-Memory Dynamic Buffers
When a port is saturated, incoming packets are temporarily stored by the switches in the bare metal network into dynamic-size buffers on high-performance memory. This minimizes packet drop rates and improves network performance.
Faster Is Better
One nanosecond is to one second as one second is to 31 years.
Minimum Comms Latency
Sending fewer larger packets means compounded communication latency is minimized. That is why storage traffic in the bare metal network is transferred as jumbo frames with MTU 9000, the largest possible packet size.
10x Extra Capacity
The standard level of overprovisioned capacity in the bare metal network is 10 times the average usage from the previous month. Capacity is increased continuously to beat this target, and we’ve never had a network congestion.
Bare Metal ID Pathfinder
IP Pathfinder is the automated IP management service that ensures the vertical scaling,
the horizontal scaling and that the instance migrations work seamlessly.
- Public IPv4 subnets are automatically allocated for each bare metal compute instance during the boot process.
- IPv6 addresses are allocated per each bare metal compute instance - making every infrastructure in the bare metal cloud IPv6 ready.
- IPv4 and IPv6 addresses allocated to the bare metal compute instances are transported to any physical server or drive allocated to the instance, during scaling or migrations.
- New subnets are added when existing classes IPs are exhausted