Evaluating a Hosting Provider’s Network Connectivity: Transport & Transit
Let’s start this topic off with some basic definitions:
- Transport – The underlying technology that is used to connect a particular node a network. Some examples of transport networks are Metro Ethernet, xWDM Waves, Wireless, Microwave, Satellite Uplink, T1/T3, DS3, OC12-192, ATM, Dark Fiber, etc.
- Redundant Physical Entrances: Separate physical networks entrances into a data center or build premises from the street.
- IP Transit – Connectivity to the larger Internet via a particular Internet provider.
- Multi-homed: A network that is connected to two or more IP Transit providers.
- BGP4: A routing protocol used to exchange routing information across the Internet. BGP4 makes it possible, and is required, in order for a hosting company or ISP to connect to multiple IP Transit providers.
- Single Point of Failure: A part of a system that will stop the entire system from working, should it fail.
- Redundant Network: A marketing buzzword that is used to dupe, deceive and otherwise mislead you into buying services from a particular provider.
I hope you can see from my list of terms that when evaluating a hosting provider’s connectivity, it is important to check many aspects. In particular, it is prudent to decide if a provider has multiple transport facilities, multiple IP Transit provider, is running BGP4, has redundant physical entrances and has no single points of failure along their entire data path from your server to their external network connectivity. Now, let’s dive deeper into each specific area and explore where typical design flaws and misrepresentations are made.
Transport – Transport is typically relatively expensive. It is the underlying ‘connection’ that always your data center or hosting company to connect to the Internet. A good real world example of this construct is power that’s running into your home. While the power company has many power companies, there is only one power line coming into your house. Many hosting companies and data centers run on the same premise — they connect with many IP Transit providers, but only have one underlying transport connection. A well run network will have physically diverse paths into their data center to make sure that physical problems, like backhoes digging up fiber, do not result in a service outage.
The very tricky thing about network transport is that there’s not much you can do to verify it is actually redundant, short of looking at the actual physical entrances into the building or looking at your provider’s invoices. Most providers will gladly show you their network; good luck getting a peek at their transport invoices, even under an NDA.
IP Transit – There are many, many companies that offer IP Transit. In fact, there’s so many companies that there are categories of companies. Tier 1, Tier 2, Tier 3 and so on. The technical definition of a Tier 1 network is a bit obtuse, but the overall point is that anyone operating a Tier 1 network is operating a very large network that has direct connectivity to many or all parts of the world. Tier 2 networks connect directly to Tier 1 providers. In many cases, Tier 2 providers can actually enhance or give further value to Tier 1 IP transport by providing intelligent network routing, DDoS protection or other managed services that are not offered by the Tier 1 provider. Every data center or hosting company should have IP transit connections to at least two providers. There are many reasons why multiple connections are better than one. From an availability perspective, if your provider only has a single connection to the Internet, it only takes one error for their IP Transit provider to take you completely offline. Thus, two IP transit providers are the smallest you should look for, when evaluating a hosting company or data center. However, knowledgeable hosting providers and data centers will also recognize that even very large Tier 1 networks have certain regions where they do well or poorly. A well run network will be connected to multiple networks that offer the best access to Asia, Europe, North America, South America and the South Pacific/Australia. Such networks will also be running BGP4, and more than likely a routing optimization platform, such as InterNAP’s Flow Control Platform.
If you want to check on your provider, you can confirm they are multihomed, running BGP4 using ‘looking glass’ tools or BGP Toolkits:
And here’s the scary part — even if a hosting company or data center has diverse transport, multiple IP Transit providers, runs BGP4, they might bring these connections back to a single switch or router. Yikes!