Featured Post

The great debacle of healthcare.gov

This is the first time in history when the president of the United States of America, or probably for any head of state around the world,...

Tuesday, June 30, 2015

Flow control & Congestion control: The two most important features of TCP that keeps the Internet alive

Flow control is the mechanism where the sender and receiver sync up the data rate between them to not to overwhelm the receiver, in the case where the receiver has less capacity than the sender.

Congestion control, on the other hand, is the sender trying to figure out what the network is able to handle. This the mechanism at the sender’s end to determine through the data loss on the transmission link and adjust the throttle accordingly to be most efficient.

Both flow and congestion controls are necessary to effectively transmit data from sender to receiver. Without flow control, the sender would overwhelm the receiver’s buffer and the data would be discarded and also sender would be forced to continuously re-transmit the unacknowledged data. This would tremendously impact the performance of the TCP protocol, throughput and the performance of the network link. Similarly congestion control helps the sender to determine if the data being sent over the network are capable of delivery to the receiver or not. There could be situation where both the sender and receiver are perfectly fine to accept a higher data rate but if the link in between isn’t capable enough, then a lot of bandwidth would be wasted just to re-transmit the lost data on the link. This would effectively make the data transmission slower than the true capacity of the link. Though both are necessary to achieve the optimal performance but they are essentially two different things:
  • Flow control is between sender and receiver, whereas congestion control is between the sender and the network
  • Flow control is dictated mostly by the receiver through negotiation, whereas the sender dictates the congestion control
  • Flow control is to sync up the data transmission between sender and receiver, whereas congestion control is to sync up the data transmission between the sender and the network link
  • Flow control is end to end but congestion control is not end to end but resides at the sender’s end alone

Implementation of flow control: TCP uses sliding window model to implement flow control. This is achieved through the use of advertised window size from the receiver. The receiver communicates the buffer size during the connection establishment and can change it anytime during the life cycle of the connection. The receiver and sender negotiates the buffer size where the SWS (Sender Window Size) is set to RWS (Receiver Window Size) that ensures that the Sender isn't sending more data than the Receiver can receive before acknowledging them.

Implementation of congestion control: TCP probes the network by starting with small amount of data to come up with optimal transmission rate in the sliding window model. TCP uses a new variable in the sliding window called congestion window, to control the streaming rate of bytes. In conjunction with sliding window’s advertised window size, this congestion window helps to determine the maximum size of the allowed window, which is the minimum of those two windows. Unlike advertised window, the congestion window is determined by the sender, determined by the network link’s ability on the data transmission. The loss of data is used as indication of congestion on the link and set the congestion window accordingly. TCP considers that the network is otherwise reliable (wireless is handled differently through).

There are various techniques that are used to implement congestion control: Slow start, Additive Increase/Multiplicative Decrease (AIMD), Fast re-transmit, Fast recovery etc. In AIMD, TCP starts streaming bytes at a minimum rate and increase the rate in an additive fashion. Another implementation is to use slow start with a small amount and then increase the rate exponentially up to the congestion threshold level. After that it goes back to additive increase until congestion is sensed, which triggers the TCP to sharply decrease the rate and also reset the congestion threshold to a lower number (depending on the implementation). This continues throughout the life cycle of the connection and sync up with the network link’s ability to handle the transmission

No comments: