There are multiple methods to identify the number of classes you should have in your QOS policy. I like to base my policies on the number of hardware queue’s my networking equipment has. Most Cisco Catalyst switches have two to four hardware queues. Some of the newer products may have more. Depending on the vendor, you may get up to eight or more hardware queue’s. Brocade uses eight queue’s on the majority of their switch ports.
Knowing the number of hardware egress queue’s that your equipment has is very important. These hardware queue’s use FIFO and no longer look at the COS, DSCP markings when a packet arrives into the queue. Due to this, an EF packet may get sent after an AF13 packet if the AF13 packet arrives into the queue first. More hardware queue’s in your hardware is better.
By default the network gear is configured with default COS or DSCP mapped to specific queue’s. You need to look at this to see if you like the way these are mapped. By default you may have EF and AF4 in the same hardware queue on a switch that only has two hardware queue’s. Maybe you want to split them up so EF goes in one queue and AF4 goes to the other queue. You need to figure out what you think will work the best to fit your needs.
Due to the different features per platform or line card, I recommend creating a policy with no more then the maximum number of hardware queue’s that is used the most. For example, the most popular switch may be a Cisco 4500, it has four queue’s. Write your policy with four classes. If the majority of your switches are Brocade switches with eight queue’s, use eight classes in your policy. More hardware queue’s the better. The number of egress queue’s should be a factor in your hardware selection.
Cisco Routers only have one hardware queue. All of the classes in your policy create a software queue. To make life simple, I recommend using the same policy on the LAN as the WAN, if possible. Usually it needs to be a little different depending on the needs of the business.
What method do you use?