Method and system for data packet queue recovery

• US 9,037,706 B2
• Filed: 05/02/2013
• Issued: 05/19/2015
• Est. Priority Date: 11/02/2010
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Net Power & Light, Inc (Amazon.com, Inc.)
• Inventors: Dubovik, Gleb, Shtayura, Vadim, Surin, Nikolay
• Primary Examiner(s): Chan, Wing F
• Assistant Examiner(s): Thieu, Benjamin M
• Application Number:
  US13/886,064
• Publication Number:
  US 20130297784A1
• Time in Patent Office: 747 Days
• Field of Search: 709/217, 709/219, 709/232, 709/224, 709/226, 709/227, 709/229, 709/250, 370/229, 370/235
• US Class Current: 709/224
• CPC Class Codes: H04L 12/6418 Hybrid transport
• CPC Class Codes: H04L 43/0852 Delays

Abstract

Resource-aware dynamic bandwidth control uses information about current network state and receiver performance to avoid, minimize and/or recover from the effects of network spikes and data processing spikes. Linear models may be used to estimate a time required to process data packets in a data processing queue, and are thus useful to determine whether a data processing spike is occurring. When a data processing spike occurs, an alarm may be sent from a client to a server notifying the server that the client must drop packets. In response, the server can encode and transmit an independent packet suitable for replacing the queued data packets which can then be dropped by the client and the independent packet present to the processor instead.

Claims

1. A processor-based method comprising:

receiving, from a server, a data stream including normal data packets (N-packets) at a client device;

queuing the N-packets in a data processing queue for processing at the client device;

estimating data packet processing latency of the N-packets;

when the data packet processing latency corresponds to a data processing latency spike, sending an alarm signal from the client device to the server; and

dropping one or more remaining packets of the N-packets in the data processing queue in response to receiving an independent network packet (I-packet) from the server;

wherein the I-packet is independently processed at the client device asynchronous to any previously received network packets;

wherein estimating the data packet processing latency includes estimating a data packet processing time for the data processing queue;

wherein the data packet processing time is estimated using a linear model; and

wherein the data packet processing time is computed as;

T=C+S*B,

where T is the estimated data packet processing time, C is a constant processing delay time independent of data packet size, B is a processing bandwidth, and S is a measure of packet size.

2. The processor-based method as recited in claim 1, wherein dropping the one or more remaining packets includes replacing the dropped one or more remaining packets with the I-packet.