CODEWITHSUNDEEP
network-programmingdnsudpprotocols
ahm5
ahm5

Reputation: 683

Why is UDP used for DNS requests and not TCP?

Why is UDP usually used for DNS requests instead of TCP?

I know that we could use TCP, but why UDP is the default protocol? Are there any reasons for that, or it is just for design purposes?

Upvotes: 5

Views: 2683

Answers (2)

Malt
Malt

Reputation: 30315

First of all, it is important to note that TCP can also be used for DNS. In practice, most DNS servers support both UDP and TCP, though TCP is rarely used for simple DNS queries and is reserved mainly for operations like zone transfers.

The biggest advantage to using UDP is the performance boost. There are several reasons why TCP DNS queries are slower:

  • TCP requires a connection to be established before each request, then subsequently torn down. So if it takes 20ms for a message to travel from your computer to the server and back (a time known as RTT - Round Trip Time), then a TCP query would require 3xRTT (60ms) to be fully processed - 20ms for opening the connection, 20 more ms for the query, and another 20ms to tear it down. UDP would only require one RTT, so 20ms.

  • Due to TCP's connection-oriented nature, more resource are needed per-connection to store and manage TCP's state. TCP requires both the client and server to have a separate socket for each and every connection.

  • UDP makes it easy to deploy anycast DNS servers. In anycast, several servers (possibly around the world) share a single IP address - e.g. 1.1.1.1. When you send a query to 1.1.1.1, one of these servers (probably one of the closest ones geographically) gets it. Since TCP involves multiple packets sent back and forth, reliable anycast is harder to achieve since you need to make sure that the packets always reach the same exact server. Otherwise, they might end up reaching different servers which won't know what to do with them.

  • Lower data overhead - a UDP header is tiny compared to the header TCP sends for every segment. Using UDP means sending fewer bytes.

  • Simplicity - UDP is a lot simpler than TCP. TCP is optimized for long data transfers and has a bunch of complex mechanisms such as flow control and congestion control for optimizing the rate of data flow. DNS doesn't need any of these mechanisms for simple queries since the typical amount of sent data is tiny.

Upvotes: 2

Patrick Mevzek
Patrick Mevzek

Reputation: 12535

UDP is default protocol because in most cases, and when DNS was designed, an exchange is a single question/response, each part fitting into a small 512 bytes packet, so there is no need to establish a long running connection, where TCP needs first a 3-way handshake before exchanging any data.

Hence in most cases UDP gives better performances and DNS is time sensitive.

But then of course UDP is easier to spoof than TCP and bigger packets can be a problem.

Upvotes: 5

Related Questions