Choosing an OS

For this cluster, we will use FreeBSD 4.9. Why? Simply because. This is not because FreeBSD is better than Linux is better than Solaris is better the Joe's OS, and so on. Honestly, any of these Unix and Unix-like systems should work like a charm. So, why do we choose FreeBSD as an example?

• It is a popular platform on which to run Squid, so community support for Squid on this platform will likely be of a higher caliber. Additionally, Squid supports the kqueue event system on FreeBSD allowing it to boast higher concurrency at lower CPU utilization than some other platforms.

• Wackamole supports FreeBSD more thoroughly than Solaris or Linux. More tested installations means less likelihood of failure.

• For the sake of an example in literature, the whole "which Linux distribution" religious war can be avoided.

We install six commodity servers with a FreeBSD 4.9 default install and assign each its public management IP address, 192.0.2.11 through 192.0.2.16, and name them image-0-1 through image-0-6, respectively.

Appropriate firewall rules should be set to protect these machines. Although this is outside the scope of this book, we can discuss the traffic we expect to originate from and terminate at each box.

• Inbound port 80 TCP and the associated established TCP session (for web serving)

• Inbound port 22 TCP and the associated established TCP session (ssh for administration)

• Outbound port 53 UDP and the corresponding responses (for DNS lookups)

• Inbound/Outbound Port 3777 UDP/TCP and 3778 UDP (to peer machines for Spread needed by Wackamole)

It is important to note that the six IP addresses above the management IPs and all client-originating web traffic will arrive to the IP published through DNS for this service.