Without SKB priority queue, queue length limits must be imposed on individual sub-queues, and there is no straightforward way to enforce a global queue length limit across all priorities. SKBprio queue enforces a global queue length limit while not restricting the lengths of individual sub-queues.
While SKB Priority Queue is agnostic to how skb->priority is assigned. A typical use case is to copy the 6-bit DS field of IPv4 and IPv6 packets using tc-skbedit(8). If skb->priority is greater or equal to 64, the priority is assumed to be 63. Priorities less than 64 are taken at face value.
SKB Priority Queue enables routers to locally decide which packets to drop under a DoS attack. Priorities should be assigned to packets such that the higher the priority, the more expected behavior a source shows. So sources have an incentive to play by the rules.
Skbprio maintains 64 lists (priorities go from 0 to 63). When a packet is enqueued, it gets inserted at the tail of its priority list. When a packet needs to be sent out to the network, it is taken from the head of the highest priority list. When the queue is full, the packet at the tail of the lowest priority list is dropped to serve the ingress packet - if it is of higher priority, otherwise the ingress packet is dropped. This algorithm allocates as much bandwidth as possible to high priority packets, while only servicing low priority packets when there is enough bandwidth.
This manpage maintained by Bert Hubert <ahu@ds9a.nl>