Doubt regarding Winsock Kernel Buffer and Nagle al

While reading this article, I got a doubt.

I understood that while trasferring small data, Nagle algorithm is enabled by default which coalesces small packets. This results in caching some data before transmission. I believe that Winsock Kernel Buffer is the place where caching happens. Correct me if I am wrong.

Does it mean that if the Winsock Kernel Buffer is set to zero with SO_SNDBUF option, will the Nagle algorithm be disabled?

If not then where does WINSOCK cache the small data?

回答1:

The KB article you refer gives your answers in this way...

To optimize performance at the application layer, Winsock copies data buffers from application send calls to a Winsock kernel buffer. Then, the stack uses its own heuristics (such as Nagle algorithm) to determine when to actually put the packet on the wire.

and, setting TCP_NODELAY or SO_SNDBUF=0 will disable Nagle algorithm as below,

The TCP_NODELAY socket option is applied to disable the Nagle algorithm so that the small data packets are delivered to the remote host without delay.

You can change the amount of Winsock kernel buffer allocated to the socket using the SO_SNDBUF option (it is 8K by default). If necessary, Winsock can buffer significantly more than the SO_SNDBUF buffer size. In most cases, the send completion in the application only indicates the data buffer in an application send call is copied to the Winsock kernel buffer and does not indicate that the data has hit the network medium. The only exception is when you disable the Winsock buffering by setting SO_SNDBUF to 0.

Reading your comment below, I realize you might be confused because setting TCP_NODELAY or setting SO_SNDBUF=0 both seem to be doing the same thing. If that is the case, please note that Nagle is applicable only over TCP streams (which segments data into packets), whereas SO_SNDBUF is a also applicable to UDP sockets.

Setting SO_SNDBUF to zero explicitly stops all output buffering and an immediate dispatch is attempted for each 'write' on the socket (at least in normal socket implementations).

Setting TCP_NODELAY will explicitly stop Nagle algorithm on TCP sockets though the send buffer may be available and used for delayed dispatch (after send success is acknowledged to the application).

回答2:

SO_SNDBUF set to 0 does NOT force an immediate send on the wire.

回答3:

Setting SO_SNDBUF to zero will not implicitly disable nagle; the nagle state maintained by the WSK is independent of where the buffer is located. It's your responsibility to keep the buffers you post valid till the transport consumes it.