4

This is just a curiosity-based question, but I might learn something useful.

With my Node.js server, as I received data through net.Server, I had it print out the size (in bytes) of each data "packet".

socket.on 'data', (data) -> console.log data.length

I noticed that most of the time it is 1374 bytes. All other times it is multiples of 1374. The highest I got out of about 200 data events was 17,862.

Where does this 1374 number come from? And why is the data length occasionally multiples of it?

My best guess is that with TCP, 1500 bytes is the most common MTU for Ethernet, and 126 other bytes make up the header of the TCP packet. Node.js may sometimes clump these packets together if it receives them quickly enough, which is why sometimes they arrive in multiples.

Nick
  • 5,228
  • 9
  • 40
  • 69

1 Answers1

1

Part of the 126 is due to the TCP header itself, which is 20 bytes.

The data is additionally padded by the "header" imposed by the server library, which is used for that very reason you've described: "clumping" up the data when multiple packets are received close enough together. When the packets are in quick succession, the extra information that is included is used to determine proper packet order and concatenate the data for the return.

This is a common technique for reducing the amount of processing that goes into each full set of data (why run processing three times on three separate pieces of data, when it can be performed once for a larger existing set), but in cases where window-negotiation is involved, it can contribute to what is known as "Silly window syndrome" (window is reduced to the point that the data being delivered is smaller than the header itself, making transmission extremely inefficient). If you've reached this issue, though, you should probably reconsider the way you are sending and receiving your data.

Aejay
  • 909
  • 9
  • 19