This topic floats around the hosting world. Rarely do you see a solution provider adequately define their levels of redundancy, including areas where they lack redundancy. As part of our rebranding and transparent client discovery, it helps to understand how a carrier is delivering redundancy, from the network edge to the server chassis.
Every carrier has a single point of failure. A “zero downtime network” can be achieved with or without single points of failure. In order to fully understand redundancy, we need to break it down into various categories and then analyze each category carefully.
Edge Network
This is the point where you receive connectivity from upstream providers. Some hosting companies rely on a single-homed (1 Carrier) connection, while other companies rely on several distinct carriers (multi-homed) and utilize a routing protocol such as BGP to handle the distribution of traffic across the various carriers.
As an example, Network Redux utilizes connectivity from Time Warner Telecom Level 3 Communications and a backup link from Integra Telecom. BGP provides redundancy between the two primary upstream providers, should one fail, traffic and routes will divert to the operational link.
Distribution Network
This is the layer which takes traffic from the edge network and distributes it to your server farms, or customer cages. This switching platform usually takes the form of a single, highly redundant chassis (e.g. Cisco 6509) or multiple switches utilizing a protocol such as VRRP for delivering redundant connectivity via multiple switches in a distribution layer.
Network Redux utilizes a multiple switch approach, and combined with VRRP provides our access layer (server switches) with active/active uplink connections.
Access Network
In general terms, this is what your stuff connects to. Servers, power units, storage area networks… utilizing VLANs traffic can be segmented as needed, and servers connect to these access layer switches for both public (to the distribution and edge) as well as private (internal traffic to other servers or backup networks).
Network Redux utilizes segmented access layer environments, each with its own active/active pair of fiber uplinks to our distribution network. We typically will pair a set of access switches and distribution uplinks with 2 42U cabinets.
Servers
A critical layer, often handled poorly by hosting providers. Servers can be built in many forms, provided by many vendors.
Network Redux utilizes Dell PowerEdge server, our current builds are deployed as follows:
- Dell PowerEdge 2950
- 8x 2.0Ghz Xeon Cores
- 32GB Memory
- 6×450GB SAS-15K RAID-10 Storage (Hot Swappable)
- Dual Redundant Hot Swappable Power Supplies
- 2-4x GigE Nics
Network Redux offsets disk redundancy concerns with the usage of server grade SAS disks in RAID-10 configurations. We offset our power redundancy concerns with dual power supplies, each which connects to a separate power distribution unit (PDU) fed by independent power sources. Yes, this means that we are paying for twice the power we actually use for failover purposes, and each server is receiving power from separate sources in the physical building.
And we have also began to offset our network card redundancy concerns by using link aggregation at our access switch levels (802.3ad) where we have two gigE links working together as a pair.
Several key items such as power redundancy and backups still need to be discussed. We will save those for another post.
At the end of the day we pride ourselves in the infrastructure we build for our customers. We do pay a premium for vendor grade and certified equipment as well as exceptional redundancy levels from Edge Network to Server Chassis.
Cheers!
