Communication Networks/TCP and UDP Protocols/Congestion Control
Congestion
editIntroduction
Congestion occurs when the source sends more packets than the destination can handle. When this congestion occurs performance will degrade. Congestion occurs when these buffers gets filled on the destination side. The packets are normally temporarily stored in the buffers of the source and the destination before forwarding it to their upper layers.
What is Congestion?
Let us assume we are watching the destination. If the source sends more number of packets than the destination buffer can handle, then this congestion occurs. When congestion occurs, the destination has only two options with the arriving packets, to drop it or keep it. If the destination drops the new arriving packets and keeps the old packets then this mechanism is called `Y’ model. If the destination drops the old packets and fills them with new packet, then this mechanism is called Milk model. In both the cases packets are dropped. Two common ways to detect congestion are timeout and duplicate acknowledgement.
Congestion control
Congestion control can be used to calculate the amount of data the sender can send to the destination on the network. Determining the amount of data is not easy, as the bandwidth changes from time to time, the connections get connected and disconnected. Based on these factors the sender should be able to adjust the traffic. TCP congestion control algorithms are used to detect and control congestion. The following are the congestion algorithms we will be discussing.
- Additive Increase/ Multiplicative Decrease.
- Slow Start
- Congestion Avoidance
- Fast Retransmit
- Fast recovery
Additive Increase / Multiplicative Decrease
This algorithm is used on the sender side of the network. The congestion window SSIZE is the amount of data the sender can send into the network before receiving the ACK. Advertised window RSIZE is the amount of data the receiver side can receive on the network. The TCP source set the congestion window based on the level of congestion on the network. This is done by decreasing the congestion window when the congestion increases and increases the congestion window if the congestion decreases. This mechanism is commonly called as Additive Increase/ Multiplicative Decrease.
The source determines the congestion based on packet loss. The packet loss is determined when the timeout happens. The source waits until the timeout time for the acknowledge to arrive. In normal cases packets are not lost, so the source assumes congestion has occurred when timeout happens. Whenever the timeout happens the source sets the SSIZE to half of the previous value. This mechanism is called Multiplicative Decrease. If timeout happens continuously, the window size is decreased until the size becomes 1. This is because the minimum value for congestion window is 1. When the sender determines that congestion has not happened, it increases the congestion window by one. This increase happens after every successful ACK received by the sender as shown below. File:Congestion1.jpg
Slow start
The main disadvantage in the Additive Increase/ Multiplicative Decrease method is the sender decreases the congestion by half when it detects congestion and increase only by one for each successful ACK received. If the window size is large and/or the congestion window size is increased from 1, then we waste many congestion windows. The slow start algorithm is used to solve this problem of increment by one. The SSIZE is the amount of data the sender can send into the network before receiving the ACK. RSIZE is the amount of data the receiver side can receive on the network. The SSTHOLD is the slow start threshold used to control the amount of data flow on the network. The slow start algorithm is used when the SSIZE is less than the threshold SSTHOLD. In the beginning the sender does not know how much data to send. It has to find how much data to send. Initially the SSIZE much be less than or equal to 2*SMSS bytes and must not be more than 2 segments. As the packets are sent the SSIZE is increased exponentially until SSIZE become greater than SSTHOLD or when congestion is detected.
When the sender detects congestion, then it decreases the congestion window by half of the previous value. Again, the slow start algorithm is used for increasing the congestion window.
Congestion avoidance
The SIZE is the amount of data the sender can send into the network before receiving the ACK. RSIZE is the amount of data the receiver side can receive on the network. The SSTHOLD is the slow start threshold used to control the amount of data flow on the network. The congestion avoidance algorithm is used when the SSIZE is greater than the threshold SSTHOLD. As the packets are sent the SSIZE is increased by one full size segment per roundtrip rime. This continues until congestion is detected.
Fast retransmission
Both the above three algorithms use timeout for detecting the congestion. The disadvantage here is the sender need to wait for the timeout to happen. To improve the congestion detection the sender uses duplicate ACK. Every time a packet arrives at the receiving side, the receiver sends an ACK to the sender. When a packet arrives out of order at the receiving side, TCP cannot yet acknowledge the data the packet contains because the earlier packet has not yet arrived. The receiver sends the same ACK which it sent last time resulting in duplicate ACK. This is illustrated below.
From the senders point of view Duplicate ACKs can arise from number of network problems. The sender cannot assume the packet sent was lost, the Duplicate ACKs may be triggered by reorder the segments, Replication of the ACK or segment. So the sender waits for 3 duplicate ACKs to determine the packet loss. TCP performs a retransmission of what appears to be the missing segment, without waiting for the retransmission timer to expire.
Fast recovery
Fast recovery algorithm governs the transmission of new data until a non-duplicate ACK arrives. The reason for not performing slow start is that the receipt of the duplicate ACKs not only indicates that a segment has been lost, but also that segments are most likely leaving the network The fast retransmit and fast recovery algorithms are usually implemented together as follows. 1. When the third duplicate ACK is received, set STHOLD no more than STHOLD = max (FlightSize / 2, 2*SMSS), where FlightSize is the amount of outstanding data in the network 2. Retransmit the lost segment and set SSIZE to STHOLD plus 3*SMSS. This artificially "inflates" the congestion window by the number of segments (three) that have left the network and which the receiver has buffered. 3. For each additional duplicate ACK received, increment SSIZE by SMSS. This artificially inflates the congestion window in order to reflect the additional segment that has left the network. 4. Transmit a segment, if allowed by the new value of SSIZE and the receiver's advertised window. 5. When the next ACK arrives that acknowledges new data, set SSIZE to STHOLD (the value set in step 1). This is termed "deflating" the window. This ACK should be the acknowledgment elicited by the retransmission from step 1, one RTT after the retransmission (though it may arrive sooner in the presence of significant out-of-order delivery of data segments at the receiver). Additionally, this ACK should acknowledge all the intermediate segments sent between the lost segment and the receipt of the third duplicate ACK, if none of these were lost.
FAQ
What causes this congestion? Congestion occurs when the source sends more packets than the destination can handle. When this congestion occurs performance will degrade. Congestion occurs when these buffers gets filled on the destination side. The packets are normally temporarily stored in the buffers of the source and the destination before forwarding it to their upper layers. Let us assume we are watching the destination. If the source sends more number of packets than the destination buffer can handle, then this congestion occurs.
What happens when congestion occurs? When congestion occurs, the destination has only two options with the arriving packets, to drop it or keep it. If the destination drops the new arriving packets and keeps the old packets then this mechanism is called `Y’ model. If the destination drops the old packets and fills them with new packet, then this mechanism is called Milk model. In both the cases packets are dropped
How do you detect congestion? Two common ways to detect congestion are timeout and duplicate acknowledgement.