At the most basic level, a switch separates a network into segments. This helps to manage network traffic by passing traffic at wire speeds to only the segment(s) that host destination nodes. But network needs vary dramatically, and switches have proliferated in size and features -- often adding intelligence that can support high-level capabilities like storage virtualization [see the Tech Closeup on switch-based storage virtualization], security and a variety of other tasks. This article outlines the major switch classes, highlights some of the most notable products in each class and offers a series of helpful acquisition and implementation guidelines.
Understanding switches
Switches are typically divided into four categories; edge, stackable, blade and director. Edge switches often appear in the SMB and take on low-end roles in the enterprise to interconnect relatively small workgroup or department-sized networks. Edge switches are noted for their low acquisition cost, simple management schemes and limited number of ports. Deployments can be scaled by simply adding more inexpensive edge switches, though scalability is limited by inter-switch link (ISL) latency and the management difficulties of numerous small devices. Today, many edge switches are reaping the benefits of features like storage area network (SAN) segmentation, security, wizard-driven management and port control. Analysts note that most companies start out with a single switch, and then grow the network by interconnecting more switches -- up to a point.
"Over time [users] realize that managing a director chassis with more ports would make their lives easier and more cost-effective," says Brian Garrett, analyst at Enterprise Strategy Group.
Stackable switches are also relatively inexpensive, but are often classified as mid-range products because of their moderate scalability. Many users start with one stackable switch and then grow their networks as needed. For example, stackable switches like the QLogic Corp.'s SANbox 5600 allows up to 16 ports to be unlocked in four-port increments using software keys. Fast ISL ports allow additional boxes to be interconnected with minimal latency. However, high-end features like hot swapping, hot code load (updates) and other maintenance features are often lacking. "I don't know that the stackable switches are there yet -- they're coming close," Garrett says. "So you might lose a little fault tolerance compared to director-class switches." Stackable switches also do not scale to the levels found in director-class models.
Blade switches provide switching capabilities in a form factor designed for blade chassis offered by manufacturers like Dell, IBM and Hewlett-Packard. By mounting a blade switch, other blade servers in the chassis can be connected through the blade switch to the rest of the network. This is cleaner and more convenient than cabling each blade server to an external switch. Blade switches typically provide the same suite of features found in edge or stackable switches, but scalability is limited, like edge switches, and they are more expensive to deploy because blade servers are not yet mainstream. And while many servers boot from the network, blade servers may encounter problems booting from the network through a blade switch. This can adversely impact blade servers, which could otherwise save space and cost by omitting local hard drives. Manufacturers are currently working to address such booting interoperability issues.
Director-class switches occupy the highest levels in a switch hierarchy. Directors feature fault tolerance, non-disruptive software upgrades, hot-swappable components, redundant power and cooling systems, and advanced security features. Ultimately, directors offer massive expandability while providing administrators with a single point of management. "It's really about scaling -- scaling the number of ports, the performance, the types of protocols, the availability/reliability and the ability to actually segregate a director into multiple virtual switches," says Greg Schulz, founder and senior analyst at the StorageIO Group. On the down side, directors have significantly higher acquisition costs than other switch classes. Since directors do not use the wizard-driven interfaces frequently found in simpler switches, management tasks can be more cumbersome and time-consuming.
Intelligence plays a greater role
It's important to point out that director switch products and intelligent switch products are not the same thing -- the terms should not be used interchangeably. "Intelligence" typically indicates the presence of advanced features in the switch, such as storage virtualization, remote mirroring, data sharing, security, protocol conversion and quality of service (QoS). While director-class switches can certainly be configured with service modules to support intelligent functions like these, it's not a necessity. "In fact, the vast majority, if not almost all mainstream director-class switches deployed, have been done so without storage virtualization blades," Schulz says. "Bottom line is that an intelligent switch does not have to be a director, and a director does not have to support intelligent or storage virtualization and storage services functions."
Even though intelligence does not have to reside in switches, the deployment of intelligent functions in the network fabric itself offers compelling benefits -- most notable is the centralization of network-based services. Rather than managing network services handled by a variety of applications running through individual servers in different locations around the network, managing an intelligent switch lowers labor costs and software licensing costs (e.g., volume managers) and improves resource utilization by reducing overprovisioning. Switch-based intelligent functions also allow heterogeneity, allowing storage managers to mix and match storage platforms and systems that might not interoperate otherwise.
The role of management tools
Any evaluation of switch products should also include a careful consideration of their management tools. Low-end switches typically provide basic interfaces with wizards that can automate many routine management tasks. Larger switches with intelligent features will offer more complex interfaces that require more time and training to master.
A wealth of different management features is available. The features vary between manufacturers, but there are some common areas to look for. For example, enhanced performance features like ISL trunking allow better inter-switch performance and port consolidation, along with performance monitoring for resource utilization analysis and service level measurement. Switch fabric management allows for configuration, management and provisioning of multiple switches from a single console -- simplifying the management of multiple switches. Health and performance monitoring tools measure operating conditions, then identify and report potential faults to the administrator or technician on duty. As more intelligence is brought to the network, security and access control features are appearing to restrict access to the fabric and authenticate administrators.
McData Corp's SANavigator Management software exemplifies a well-integrated product. The suite includes performance monitoring and trending down to each port, storage planning, event management and automated response, LUN management for storage systems from EMC, HDS and IBM, as well as security and authentication features.
Vendors and product selection
Although there are relatively few enterprise-class switch vendors, each vendor offers an array of switch products -- and the products are constantly improving to address factors like storage network complexity, burgeoning storage and server proliferation, increasing bandwidth requirements and added security concerns.
At the low-end, departmental/workgroup (aka "edge") switches are packing more features and functionality, while remaining cost-effective and easy to deploy. As one example, Cisco System's MDS 9216 fabric switches provide 16 fibre channel ports running at up to 2 Gbps, along with Gigabit Ethernet ports for FCIP or iSCSI interconnections. The 9216 includes advanced services such as access control lists (ACL), fibre channel congestion control (FCC) for traffic management, and quality of service (QoS) support. Brocade Communications Systems Inc's SilkWorm 200E provides 8-16 ports of 4 Gb per second (Gbps) Fibre Channel (FC) connectivity for SAN deployment. The unit features Fabric Shortest Path First (FSPF) traffic routing for greatest efficiency and scalability from eight to 12 or 16 ports in four-port increments. Brocade's EZSwichSetup wizard eases SAN deployment, and per-port statistics help with fast problem diagnosis and correction.
[Continued on page 2]
