Building Dreamhack, part three

DHCP design for IPv4 on Dreamhack

I will describe the protocol DHCP in general and specific the DHCP design that we use at Dreamhack for IPv4.

DHCP for IPv4
DHCP is a layer 3 protocol used for dynamic assignment of ip addresses and options to clients. The client device sends a layer 3 broadcast to 255.255.255.255 on the local network destination UDP port 67. This message is called a DHCP discovery and it is a request for a free ip with options. The server answers the broadcast on UDP port 68 with a DHCP offer. This offer contains information about IP, subnet mask, lease time, options and the ip address of the DHCP server. The client then sends a DHCP request to the DHCP server accepting the offered lease. When the server receives the DHCP request it sends back a DHCP acknowledgement with lease duration and options.
When half of the lease time has gone the client tries to renew its lease by sending a DHCP request message to the DHCP server. If the client does not get a response from the server it will continue to send DHCP request messages to the specific DHCP server on a regular interval. When the lease time ends the client will begin the process from start by sending a DHCP discover.

DHCP design at Dreamhack
At Dreamhack all the clients use DHCP for configuring IPv4, subnet mask, default gateway, SMTP, TFTP, DNS and NTP servers.

For hardware redundancy we have three DHCP servers. For operating system redundancy we run Debian and FreeBSD. We have one active/primary server that syncs its lease file to the two passive/secondary DHCP servers. If the primary goes down or a severe OS related issue occurs then we can start using one of the secondary.

DHCP monitoring and statistic
We have our own developed DHCP scope monitoring and statistic system written in ruby by me 🙂 The system has two daemons and a web application.

Daemon one tails and parses the DHCP lease file, and parses the scope information. Daemon one then sends the parsed output to MySQL and MongoDB datastores. Daemon two analyzes the data in the datastores and creates statistics and graphs. This information is then made available through a web application developed with the Sinatra framework.