Rate-optimized congestion management
AssigneePensando Systems Inc.
InventorsRong PAN, Peter NEWMAN
Life of this patent
Estimating the term…
Abstract
This disclosure provides methods and systems for reducing congestion in RoCEv2 networks. The method is configured to operate large-scale in data centers on traffic flowing from a sender node to a receiver node. The method described has three stages: a fast start stage, a transition stage, and a regulation stage. In the fast start stage, the sender sends data to the receiver at a fast initial rate. This may continue until the receiver observes a congestion event. When this happens, the sender reduces the data transfer rate as the method enters the transition stage. From a reduced rate, the method enters the regulation stage, where the rate is increased using a combination of a feedback control loop and an additive increase multiplicative decrease (AIMD) algorithm.
Claims55
A computer-implemented method of managing congestion in a network protocol, the method comprising:
- a) detecting a congestion event in the network;
- b) slowing a data transfer rate from an initial rate to a minimum rate;
- c) increasing the data transfer rate to half of the initial rate;
- d) periodically monitoring a one-way latency of the network;
- and
- e) adjusting the data transfer rate using a feedback control algorithm and an additive increase multiplicative decrease (AIMD) algorithm.
A system for managing network congestion in a network, comprising a plurality of computing nodes, a computing node comprising:
- a) a memory for storing one or more previously obtained latency values and a set of instructions;
- and
- b) one or more processors configured to execute the set of instructions to:
- i) detecting a congestion event in the network;
- ii) slowing a data transfer rate from an initial rate to a minimum rate;
- iii) increasing the data transfer rate to half of the initial rate;
- iv) periodically monitoring a one-way latency of the network;
- and
- v) adjusting the data transfer rate using a feedback control algorithm and an additive increase multiplicative decrease (AIMD) algorithm.
A computer-implemented method for managing network congestion in a network, the computer-implemented method comprising:
- a) a fast start stage, comprising setting a data rate to an initial data rate and performing data transfer between a sender and a receiver until the receiver detects a congestion event;
- b) a transition stage, comprising reducing the data rate to a minimum data rate for a period and then increasing the data rate to half of the initial rate for a period;
- and
- c) a regulation stage, comprising using a combination of a feedback control loop and an additive decrease multiplicative increase algorithm to modify the data rate.
A system for managing network congestion in a network, comprising a plurality of computing nodes, a computing node comprising:
- a) a memory for storing one or more previously obtained latency values and a set of instructions;
- and
- b) one or more processors configured to execute the set of instructions to implement:
- i) a fast start stage, comprising setting a data rate to an initial data rate and performing data transfer between a sender and a receiver until the receiver detects a congestion event;
- ii) a transition stage, comprising reducing the data rate to a minimum data rate for a period and then increasing the data rate to half of the initial rate for a period;
- and
- iii) a regulation stage, comprising using a combination of a feedback control loop and an additive decrease multiplicative increase algorithm to modify the data rate.
Dependencies are inferred from the claim text.