In a connection between a server and a client; both of them share the maximum number of bytes they can receive at a time. This maximum number of bytes is called the Window size. Client's receive window size will be the server's send window size and vice versa.
Internally, the receive window size of a client or a server is the size of the receive buffer for that connection in the network device in it.
In other way, a window size is the amount of data that can be sent through a connection till the receipt of acknowledgement for the sent data.
TCP uses this window concept to achieve good performance and to avoid congestion (bottleneck). A congestion can result in a case where the sender sends data in a higher rate than the receiving rate of the other end (sender window > receiver window).
In most operating systems, TCP send and receive socket buffer sizes are set before setting up a connection. In UNIX and Linux, for example, they are set through sysctls.
Finding appropriate Window size
Theoretically, the window size is the bandwidth delay product calculated as the product of bottleneck bandwidth and the roundtrip time.
Suppose, if we have a connection between two different networks; one is a Gigabit Ethernet (1000Mbps) and the other is a fast Ethernet (100Mbps). The bottleneck bandwidth is that of fast Ethernet.
We can find that connection's roundtrip time by running a ping. In a normal operation, the ping command sends an ICMP request to the destination and waits for the acknowledgement. The destination sends ICMP reply to the source. The ping command prints the roundtrip time for the ICMP request/reply combination. Let's assume that the roundtrip time for that connection was 20ms.
Window size = 100Mbps * 20ms
= 100 * 10^6 * 20 * 10^(-3)
= 2000 * 10^3 bits
= (2 * 10^6 )/8 bytes
= 250KB
For more details, you can refer to the below pages
Read More...
