Storage is indispensable. Without it, data centres don’t exist, business doesn’t make money and tablets have nothing to display.
But storage by itself isn’t enough. The data it holds has to be processed by applications and sent to users, customers and partners, and that means it has to flow quickly, reliably and without interruption. Storage requires a network and a transport protocol to send and receive the data residing in storage pools.
But not just any old network will do. Today, there are a number of storage protocols to choose from, including Fibre Channel, iSCSI, NFS, CIFS, and the newcomer, FCoE. Data centres typically use more than one protocol, but regardless of which these may be, the storage network has to meet the same five critical requirements. These are resiliency, high availability, low latency, equal cost multi-path routing (ECMP) and consistent policy enforcement.
Every storage network has to provide these or data gets corrupted, applications crash and customers go to a competitor instead of waiting for online order screens to refresh. With a client/server application, best effort is almost acceptable, but with storage traffic, guaranteed delivery is essential. Ensuring performance and guaranteed delivery in a network isn’t easy, but it is essential for storage traffic.
A fabric network provides all five of these critical requirements for any storage protocol. Fabrics are not hierarchical “trees”, so all nodes are equidistant from each other in terms of bandwidth and latency. A unique property of a fabric is the ability to cross-connect all nodes using a flat topology. Every node can talk to any other node without loss of bandwidth or variations in latency. Fabrics are the perfect way to transport storage traffic, as storage protocols are very sensitive to congestion delays and variable latency.
Another important property of a fabric is resiliency to disruption and path failure. Just as a tear in a piece of cloth stays localised and doesn’t destroy the entire cloth, loss of a single link, switch port, or switch does not affect traffic flowing through the rest of a fabric. But unlike your jeans, fabrics are self-healing. They detect path failures and automatically re-route traffic around the failure to the shortest remaining paths without applications or users being aware of it. This self-healing property is very important for storage traffic because server applications and databases expect reliable transport of storage data even when a path fails. And with the growth of server virtualisation, more applications are running on a single server, so self-healing fabrics are more important than ever.
Finally, data security and integrity rely on policies to define access control so applications access only their own storage extents within a storage pool. Ensuring policies are consistent on every port in every switch in a fabric can be time consuming and error prone, particularly as the size of the network increases. With server virtualisation and live virtual machine migration, manual policy management becomes a nightmare for the network administrator. Fabrics solve this problem by abstracting policies from the physical ports and switches in a fabric. Policies are applied one time to the fabric rather than the physical switch ports. The fabric control plane automatically enforces the policy for all switches and ports in the fabric so data never gets corrupted, goes missing or is compromised.
With fabric technologies, IT departments can confidently use any storage protocol they desire with the assurance that all five critical storage requirements – resiliency, high availability, low latency ECMP routing and security – are being handled. Fabrics make sure storage administrators stay off the administrative treadmill as storage continues to grow.