Most common queries are translation of a hostname to an IP address. It is also possible to request additional information from DNS. Queries are mediated by a resolver. The resolver is a DNS client that asks the name server. Because the database is distributed worldwide, the nearest name server does not need to know the final response and can ask other name servers for help. The name server, as an answer to the resolver, then returns the acquired translation or returns a negative answer. All communication consists of queries and answers.
The name server searches in its cache memory for the data for the zone it administers during its start. The primary name server reads data from the local disk; the secondary name server acquires data from the primary name server by a query zone transfer of the administered zones and also saves them into the cache memory. The data stored within the primary and secondary name servers is called authoritative data. Furthermore, the name server reads from its memory cache/hint the zone data, which is not part of the data from its administered zone (local disk), but nonetheless enables this data to connect with the root name servers. This data is called nonauthoritative data. In the terminology of BIND program version 8 and 9, we sometimes do not speak of them as primary and secondary servers, but as master servers and slave servers.
Figure 1.4: Primary name server loads data from a disk, while the secondary server acquires data by zone transfer query
Name servers save into their cache memory positive (and sometimes even negative) answers to queries that other name servers have to ask for. From the point of view of our name server, this data acquired from other name servers is also non-authoritative, thereby saving time when processing repeated queries.
Figure 1.5: Stub resolvers and caching resolvers
Requirements for translations occur in a user program. The user program asks a component within the operating system, which is called a resolver, for a translation. The resolver transfers the query for translation to a name server. In smaller systems, there is usually only a stub resolver. In such cases, the resolver transfers all requirements by DNS protocol to a name server running on another computer (see Figure 1.5). A resolver without cache memory is called a stub resolver. It is possible to establish cache memory for a resolver even in Windows 2000, Windows XP, etc. This service in Windows is called DNS Client. (I think this is a little bit misleading as a stub resolver is not a proper DNS client!)
Some computers run only a resolver (either stub or caching); others run both a resolver and a name server. Nowadays, a wide range of combinations are possible (see Figure 1.6) but the principle remains the same:
1. The user inserts a command, then the hostname needs to be translated into an IP address (in Figure 1.6, number 1).
2. If the resolver has its own cache, it will attempt to find the result within it directly (2).
3. If the answer is not found in the resolver cache (or it is a stub), the resolver transfers the request to a name server (3).
4. The name server will look for the answer in its cache memory.
5. If the name server does not find the answer in its cache memory, it looks for help from other name servers.
6. The name server can contact more name servers by a process referred to as iteration. By iteration, the name server can access or contact a name server, which is an authority on the answer. The authoritative name server will then give a last resort answer (negatively if there is no information in DNS corresponding with the inserted name).
7. But if the process described above does not return the result fast enough, the resolver repeats its query. If there are more name servers listed in the resolver configuration, then it will send the next query to the next name server listed in the directory (i.e., another name server). The directory of name servers is processed cyclically. The cycle starts for the particular query from the name server, which is listed in the first position.
Figure 1.6: Name server and resolver
DNS uses both UDP and TCP protocols for the transport of its queries/answers. It uses port 53 for both protocols (i.e., ports 53/UDP and 53/TCP). Common queries such as the translation of a name to an IP address and vice versa are performed by UDP protocol. The length of data transported by UDP protocol is implicitly limited to 512 B (a truncation flag can be used to signal that the answer did not fit into 512 B and it is therefore necessary for the query answer to be repeated by the TCP protocol). The length of UDP packets is limited to 512 B because a fragmentation could occur for larger IP datagrams. DNS does not consider fragmentation of UDP as sensible. Queries transporting zone transfer data occur between the primary and secondary name servers and are transported by TCP protocol.
Common queries (such as the translation of a name to an IP address and vice versa) are performed with the help of datagrams in UDP protocol. The translations are required by a client (resolver) on the name server. If the name server does not know what to do, it can ask for translation (help) from other name servers. Name servers solve questions among themselves by iteration, which always starts from the root name server. More details are available in Section 1.10.
Figure 1.7: Required answer for a translation
There is a rule in the Internet that a database with data needed for translations is always saved on at least two independent computers (independent name servers). If one is unavailable, the translation can be performed by the other computer.
In general, we cannot expect that all name servers are accessible all the time. If the TCP protocol is used for a translation, attempts to establish a connection with an inaccessible name server would cause long time intervals while the TCP protocol is trying to connect. Only when this time interval is over is it possible to connect to the next name server.
The solution for this in UDP protocol is more elegant: A datagram containing a request for the translation is sent to the first server. If the answer does not come back within a short time-out interval, then a datagram with a request is sent to another name server, if the answer does not come back again, it is sent to the next one, and so on. If all possible name servers are used, it will start again from the first one, and the whole cycle repeats until the answer comes back or the set interval times out.
Round Robin is a technique that can be used to equally load several machines (Load Balancing). It is possible to use this technique for the majority of name servers (including Windows 2000/2003). This is a situation where we have more than one IP address for one name in DNS. For example, we may operate an exposed web server and because the performance of the machine is not
sufficient, we buy another or two more. We start running the web server on all three of them (for example, www.company.com). The first one has an IP address 195.1.1.1, the second one 195.1.1.2, and the third one 195.1.1.3. There will be three records in DNS for www.company.com, and each of them will have a different IP address. Round Robin technique ensures that the answer to the:
1. first query (to the first user) will be that the web server return addresses 195.1.1.1, 195.1.1.2, and 195.1.1.3
2. the answer to the next query (to the second user) will be that the server return IP addresses 195.1.1.2, 195.1.1.3, and 195.1.1.1.
3. the answer to te next query (may be 3rd user) will return IP addresses 195.1.1.3, 195.1.1.1, and 195.1.1.2.
4. procedure are repeating from 1st point again and again.