Setting up a SAN infrastructure

>

To design a scalable SAN infrastructure one must formally assess the environment the SAN is meant to support. The assessment aids in deriving strategic network characteristics thus simplifying the purchase criteria for selecting the appropriate types of switches. Besides looking at today''s business needs, try to analyze future business as well to fully utilize the SAN infrastructure.

Designing an ideal SAN topology requires analysis of how the supported applications affect the entire data environment from data consolidation to disaster recovery strategies. Key attributes that must be assessed include:

• Application prerequisites - Downtime tolerance, performance, and growth.
• Data storage requirements - location of data, and amount of data.
• Backup and disaster tolerance strategies
• Network connectivity requirements - port count requirements, network traffic patterns, and bandwidth requirements.
• Server connectivity requirements

Organizations must first choose among the several types of Fiber Channel switching products in the market. These switches can be broadly classified into three types - directors, fabric switches and arbitrated loop switches (loop switches). Remember that all vendors'' products differ; one vendor''s fabric switch could be another vendor''s director.

The several switch types differ in other areas as well - availability, scalability, cost and performance. Each switch offers a unique combination of functionality. No particular switch is therefore best across all applications.

Directors

offer an availability of less than five minutes of downtime per year. These switches provide high-performance bandwidth, owing to their higher port count and non-blocking architecture. Therefore it allows all ports to simultaneously intercommunicate without performance degradation or added latency.

Directors are implemented for applications such as:

• Mission-critical applications where there is no tolerance for downtime.
• Backbone for enterprise SANs - provides a known data-path.
• Performance-intensive applications, where any-to-any port intercommunication ought to occur at high-bandwidth speeds.

Fabric switches

provide full speed data transfer between all ports on the switch sans interference or blocking. Like directors, this switch varies with vendors. While choosing a vendor, look out for features like redundancy, port count, etc. These switches are commonly implemented as the principle building blocks of a smaller SAN or as a consolidation point for larger enterprise SANs.

Fabric switches are implemented for the following applications:

• Departmental connectivity
• Applications where distributed storage predominates

Loop switches

are connectivity devices focused upon workgroup area connectivity. It is suitable for low-bandwidth utilization devices and support of legacy arbitrated loop devices such as tape. It offers the lowest cost connectivity option. Loop switches must maintain a low port count (eight) to decrease bandwidth contention.

The above switches are implemented for the following applications:

• Low-cost fabric connectivity which is ideal for low-bandwidth devices
• NT Server consolidation
• Tape consolidation
• Legacy loop connectivity to a switched fabric

Choosing the right switch

This being the foundation, extreme care must be taken as it will determine how well the infrastructure supports the intended application. Additionally, it also ascertains its versatility as business needs evolve.

Based on availability, performance, scalability and cost, businesses must compare and choose the right switch. Keep in mind that there is no one switch which is best across all applications.

Availability

is a measurement of uptime and applies to an application''s requirements and the fundamental network. Availability of switches is expressed as a percentage average of uptime, usually ranging from a low of 99 per cent (over 3.6 days of downtime), to 99.999 per cent (less than five minutes of downtime) per year. High-availability networks are supported by directors and redundant fabric switches. Fabric and loop switches provide sufficient availability for workgroups and departmental office applications that are able to tolerate periods of downtime.

Performance

must be taken into account whether implementing a single switch or an entire fabric. The switch architecture determines the performance characteristics within the switch. Fabric performance is a consequence of the overall fabric design as well as traffic patterns through the fabric.

Scalability

pertains to the ability of the SAN to increase in size with minimal disruption to ongoing operation. Both director and fabric switches support fabric connectivity and scalability of the fabric through the addition of more switches. However, the key to maximizing scalability is to minimize the number of switches being added.

Cost

is an essential consideration here, as is with all network infrastructure purchases. Each switch type cost varies. The cost should be primarily assessed at the fabric level as total cost of ownership of the SAN, and not a particular switch, is most significant. Cost of managing the SAN should be taken into account as well. A large fabric built upon low-port count switches will have a higher associated management cost than a similar fabric built with large-port count directors or switches.

In a large fabric designed for high availability, directors will likely be more cost effective than a comparable fabric designed with fabric switches. On the other hand, a 16- or 32-port fabric switch would prove ideal in a smaller department environment. Overall, the switch cost can change considerably depending upon the implementation and the size of the SAN.

While choosing the vendor, remember that it is a complicated assignment for a single network management application to manage switches from different vendors.

Zeroing in on the right SAN infrastructure is critical as businesses can then ensure that their network can effortlessly expand beyond its initial configuration. While cost effectively it can support performance and data availability needs.