Until recently, for businesses that could justify the expense of a SAN, their SANs were mainly the domain of high-end...
servers and shared storage arrays. Director-class switches were the largest national and multi-national organisations preserve that had dedicated storage teams due to the stratospheric purchase and running costs.
Across data centres, the cost-per-port of SAN switches and fabrics has dropped as rapidly as the density of individual devices has increased. Where 64-port switches were once a rarity, they are now considered edge devices, and the largest director switches can now contain over 500 ports.
This brings us to a question – Are edge switches a thing of the past? While the average SAN is rapidly growing in size, the percentage of SAN fabrics that measure hundreds of ports still remains small.
As companies grow, so does the size of the SANs they implement and must administer. Most SAN environments begin with a pair of switches, in separate fabrics for redundancy, with more switches added when the existing devices fill up. A point will usually be reached where the administration of the fabrics becomes cumbersome due to factors such as ISL bottlenecks, too many ports being used as ISLs or host/storage placement issues. This is generally the case when more than five switches are in any one fabric. At this point a refresh is usually required and should involve a complete redesign of host and storage placement.
The standard configuration for medium-sized to large SANs is a core-edge design, which comprises large switches or directors at the core, and lower-capacity switches at the edge. Hosts are generally connected to edge switches, while storage is generally be connected to the core switches, such that a data packet will traverse a maximum of two switches in order to reach its destination. If required, multiple ISLs can be configured between the edge and core switches to increase bandwidth between hosts and storage.
While this SAN configuration is a good compromise, there are still drawbacks.
- Host and storage placement will need to be planned to avoid bottlenecks,
- Edge switches are less reliable than core switches
- Hosts will be plugged into ports that they cannot possibly utilise effectively,
- Many ports will still be needed for ISLs, and
- the fabric will still become complex to administer as it grows.
Today's intelligent director switches use blades that can aggregate bandwidth between ports, enabling administrators to increase the nominal rate for storage ports and decrease the rate for host ports. As the port count increases, blades can be added without interruption to service or reconfiguration of devices. Director switches also have no single point of failure, offer up to "five nines" availability, and can have maintenance tasks such as operating system upgrades carried out without service denial. While storage or hosts should never be added without regard to their usage, their placement is ultimately less important on a director switch due to the increased bandwidth of the backplane over traditional ISLs.
The deployment of director switch only fabrics will almost certainly be more expensive than a traditional core-edge fabric but the price gap is narrowing considerably. Over the life of the fabric, however, the ease of management and upgrade of a single director, together with reduced maintenance time and outages, may make the TCO of a director-only fabric lower. This could mean that the days of the core-edge fabric for large enterprises may be numbered.
About the author: Steve Pinder is a principal consultant at GlassHouse Technologies (UK) Ltd. He has more than 11 years experience in backup and storage technologies and has been involved in many deployments for companies of varying sizes. Prior to working for GlassHouse, Steve was an IT contractor concentrating on backup and network management roles.