Choosing the Right Network Edge
Stackable vs. Chassis Switching
By James Dunn, Solution Architect
The debate around the use of stackable vs. chassis based Ethernet switches has been around for a while. The driver for this is the blaring cost savings that could potentially be achieved through not using chassis based switches.
As we all know, upfront capital costs for a single chassis based switch can be huge and are not suitable for annually topped up network budgets that are meant for port and bandwidth upgrades. Cost is not the sole focus of this debate though. You still need a fully functioning network. Here we compare the other key influencing factors in choosing the right network edge: scalability, manageability and serviceability, availability and reliability and physicals.
Scalability
Scalability is a critical concern. Will your network investment scale effectively in areas such as port capacity and bandwidth? With chassis switches you can add port capacity by introducing more blades, providing there are enough slots available in the chassis itself. Chassis have a fixed limit on the number of slots physically available, backplane bandwidth and even the amount of power they can provide to the rest of the modules, so it is essential that thorough planning and foresight has gone into the design.
Standalone switches can be added to a stack as and when extra port density is needed in a ‘pay as you grow’ fashion, and generally have a switch limit of 8 which is standard for most vendors. Stacked switches can be thought of as a virtual chassis and for vendors like Extreme Networks, can be mixed and matched with Fast Ethernet, Gigabit Ethernet to 10 Gigabit Ethernet, in copper, Power over-Ethernet (PoE) or Fibre interfaces based on requirements. This flexibility makes for some interesting design scenarios where you can get a lot of the modular benefits of chassis based switches in a stack (apart from services modules such as firewall or content switching for example).
Manageability and Serviceability
With chassis solutions simply add blades and then manage them from the management interface card/s. Stackable switches can be managed by assigning a management IP address to the stack so both can be administered as a single entity. Any configuring performed from this management interface, is normally replicated to every switch in the stack.
Any upgrade performed to the OS (operating system) on the management card or supervisor, upgrades the entire chassis. For stackable switches, all switches participating in the stack must be running the same OS version, which may require individual switches to be upgraded to the OS version that the Master switch is using.
Availability and Reliability
These terms are often spoken in the same sentence, even though they mean different things. By using a non-redundant supervisor or management card in a chassis switch, there is the risk of the entire switch going down due to software/hardware failure. Adding another supervisor or indeed keeping a hot spare may lead to increased upfront costing.
What is interesting with stacks is that it is possible to retain functionality during failure because a single switch failure, even of the Master, will lead to a hitless failover election of the standby/backup switch in the stack, retaining as much 75% stack functionality in a stack with four switches.
Physicals
Space, power, and cabling are of equal importance, but you need to consider the following
- Space is a rare commodity and a closet cupboard more often than not serves multiple purposes, so it is simply not realistic or indeed possible to consider introducing a 5U chassis into the mix.
- If a chassis is not fully populated with line cards, then space is being wasted – which has a value attached to it nowadays. If this is a concern then it may be worth using stackable switches as they only take up space that they actually use.
- Chassis switches have excellent redundant load sharing power supply options that have significantly less cable management overhead than stackables (2 power cables, vs. up to 16 for a fully populated stack)
There are many architecture and design differences between both types of switches that are beyond the scope of this article but if you would like to discuss this, or know more about how IP Integration approach this debate, then please contact me directly at James.Dunn@ipintegration.com