Predictive management of a network buffer
AssigneeCable Television Laboratories, Inc.
InventorsGregory C. White, Joey Padden, Takashi Hayakawa
Life of this patent
Estimating the term…
Abstract
Predictive management of a network buffer is contemplated. The network buffer maybe predictively managed to control packet drop based at least in part on predicted sojourn time. The predicted sojourn time may be determined to predict time needed from an arriving packet to travel through a queue of the network buffer.
Claims20
A non-transitory computer-readable medium having a plurality of instructions executable with a processor to facilitate managing a network buffer, the network buffer having at least a first queue and a second queue, each of the first queue and the second queue correspondingly enqueuing packets at an input and to dequeuing packets at an output, the plurality of instructions being sufficient for:
- predicting sojourn time for each packet prior to being enqueued into one of the first queue and the second queue, the sojourn time estimating time needed for the corresponding packet to travel through one of the first queue and the second queue;
- and
- one of dropping and not dropping each based at least in part on the corresponding sojourn time.
A network element comprising:
- separate queues for queueing packets for subsequent transport over a link;
- a controller having non-transitory computer-readable medium with a plurality of instructions executable with a processor to facilitate to managing packet drops, the plurality of instructions being sufficient for:
- i) calculating a sojourn time for packets prior to the corresponding packets entering one of the queues, each sojourn time estimating time needed for the corresponding packets to travel through an entirety of the corresponding queue before being dequeued;
- and
- ii) one of dropping and not dropping the packets depending on the sojourn time associated therewith, including dropping the packets having the sojourn time above a corresponding sojourn threshold and not dropping the packets having the sojourn time equal to or less than the corresponding sojourn threshold.
A method for managing packet drop for a network element interfacing multiple connections with a bottleneck link, the method comprising:
- directing packets having a first marking to a first queue of the network element, the first queue having a first size;
- directing packets having a second marking to a second queue of the network element, the second queue having a second size greater than the first size;
- and
- managing packet drop for the first queue differently than packet drop for the second queue to account for a difference between the first size and the second size.
Dependencies are inferred from the claim text.