Packet Marking for Fast Data Flow Stabilization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Internet congestion control methods, particularly TCP's slow-start mechanism, lead to prolonged time to stabilize at the correct transmission rate and significant packet loss when a new flow starts or restarts, causing undue congestion and impact on other flows due to overshooting the available capacity.
Innovation Solution
The method involves using packet markings to indicate the longest instantaneous queue length, allowing the source node to rapidly adjust its transmission rate to match the available capacity without overshooting by calculating a mismatch ratio based on queue length feedback, enabling quicker stabilization and reduced packet loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If TCP uses slow-start algorithm to exponentially increase sending rate by doubling congestion window every round trip, then the flow can probe for available capacity, but it takes prolonged time to stabilize at the correct rate and causes significant packet loss due to overshooting available capacity
Solution Approach 1:
The patent applies preliminary action by having the receiver proactively measure and mark packets with queue length information before the sender can probe for capacity. This allows the sender to immediately adjust its sending rate based on pre-obtained queue length measurements, eliminating the traditional slow-start probing phase and achieving both fast stabilization and accurate capacity detection.
Solution Approach 2:
The patent implements feedback by having the receiver continuously measure queue lengths and embed these measurements in feedback packets sent to the sender. This real-time queue length feedback enables the sender to precisely adjust its sending rate to match available capacity, avoiding both overshooting and prolonged stabilization periods.
2Productivity
If TCP doubles congestion window every round trip during start-up phase, then the flow rapidly increases sending rate, but it causes undue congestion and packet loss impacting other flows sharing the same capacity
Solution Approach 1:
The patent uses real-time queue length feedback from the receiver to enable the sender to precisely match its sending rate to the available capacity. This eliminates the need for aggressive exponential probing that causes congestion, while still achieving fast convergence to full capacity utilization without harming other flows.
Solution Approach 2:
The patent changes the control parameter from congestion window size (in traditional TCP) to queue length-based rate adjustment. By using queue length as the primary control parameter, the system can rapidly adapt the sending rate to match available capacity without the overshooting behavior inherent in exponential window doubling.
3Reliability
If TCP uses conservative initial congestion window of 3-10 packets, then the flow avoids initial congestion, but it takes much longer to discover available capacity and reach optimal transmission rate
Solution Approach 1:
The patent applies preliminary action by having the receiver perform capacity measurements and prepare queue length feedback before the sender needs to increase its rate. This eliminates the need for conservative initial windows and slow probing, as the sender receives accurate capacity information immediately and can safely transmit at optimal rates from the start.
Solution Approach 2:
The patent implements continuous queue length feedback that allows the sender to rapidly and reliably discover available capacity without using conservative initial windows. The feedback mechanism provides trustworthy queue length information that enables aggressive rate increases without causing congestion, achieving both fast capacity discovery and reliable congestion avoidance.
Data Source
AI summary
Disclosed is a method and apparatus for a packet data flow to use markings on packets (93, 94) that indicate the longest instantaneous queue length on a path in order to rapidly increase the flow rate up to an available capacity of a network node without overshoot, in order to avoid the TCP Slow start process.