Reliability

We worry about outages more than anyone.

Not all data centers, networks, or services are created equally. We think of everything that could possibly go wrong. TeraSwitch is built around the idea that no link in a chain can be relied on.

Striving for Better

Our designs and guidelines have been built and tested over our nearly 20 year history as a hosting company. We're constantly changing how we operate to better meet our own reliability guidelines as technology improves.

Open book operations

TeraSwitch is proud of our design, and we don't plan on hiding it. We believe our operational practices are our best asset.

If you have any questions, concerns, or requests for more information, our engineers would be happy to elaborate for you.

Facility and Critical Load

The data center(s) are the baseline of any service. If you can’t trust the facility or room you’re in, no services from it can be relied on externally.

  • Structure: Strong facility for natural disasters, roof leaks, flood mitigation, and away from hazardous locations.
  • Power: Consistent utility power that minimizes the number of events that the UPS and generators have to make up for.
  • Cooling: Ensure cooling is more than adequate, even if half of all CRAC units fail. CRAC units and their roof-top units are powered by A+B power.
  • Vendors: The data center infrastructure is built and maintained by industry experts.
  • Expansion Space: Expanding is pre-planned, and not done in a way that risks the operation or availability of existing assets.

Networking

Without a good network, your data center might as well not exist. The design and concerns that molded its’ design are critical.

  • Pairs of everything: All routers and switches are duplicated in a manner that allows not only fault tolerance, but customers and devices to connect to physically independent network points.
  • Multi-Path: All services and network connections have multiple physical inter-connects. A single physical connection should never be able to take down a logical connection between devices.
  • Upstream: Upstream transit providers are diverse and high quality. Keeping in mind the location, peering, and shared infrastructure between providers is key. Providers should be unrelated by any common single points of failure.
  • Warranties aren't good enough: Keeping spares on-site are required to minimize periods of reduced redundancy.
  • Ready for more: All inter-connections are LACP bonds even if they're single cables. This makes expanding capacity convenient, painless, and less risky.
  • Over-built: Capacity planning is done with thinking of 5 years in the future. Ports and cables are cheap, but bottlenecks are costly.

Cloud and Services

Without a good network, your data center might as well not exist. The design and concerns that molded its’ design are critical.

  • Fully Redundant: All cloud servers and network devices use the A+B power and diverse network uplinks that are available within the TeraSwitch facilties.
  • Automated: Devices are controlled and in compliance by using automation tools to ensure conformity with expected configuration.
  • Trust but Verify: Automation tasks that are run on production service hosts are verified manually by an engineer before change operations are completed.
  • Extra Capacity: Each Cloud availability zone is ready for bulk use of additional resources, and ready for migration of VMs or data from other availability zones.
  • Separate Networking: Cloud networking for storage traffic is completely isolated on different network interfaces from VM network traffic. This allows for SAN/Ceph traffic to bypass other traffic without a dangerous storage traffic bottleneck.