The release of Ubuntu 6.06 (Dapper Drake), back in June, brought not only a new desktop system to the Linux world, but also a server system with long-term commercial support. It has one key advantage over similar offerings from Redhat and Novell; the flexibility of the Debian dpkg packaging system.
This was of particular interest to me, as a system administrator who generally installs Debian, if given a choice. One of the annoying problems with Debian has been its potentially short support lifespan; essentially as long as it takes to get two more releases out. Admittedly this hasn't been a real problem, to date, but not having firm dates has been an issue in some environments in which I've worked.
Another was its perceived lack of commercial support, which often made it very difficult to bring into a corporate environment. While I've worked in situations where I had complete authority to use whatever OS I chose, I've also been in workplaces where it has been made clear that Debian simply would not be used, due to the lack of a commercial organisation providing security support.
Ubuntu's server release solves both of these problems, so I installed a copy to see how it held up.
Quagga is a fork of the zebra routing protocol daemon, with implementations of a number of routing protocols, including RIP, RIP2, RIPNG, OSPF, OSPF6 and BGP. A fairly clever feature of Quagga is its configuration system; every daemon can as command-line interface which can be accessed via telnet, where commands can be entered to take effect immediately. Furthermore, these CLIs have been designed to mimic Cisco's IOS, so anyone with a bit of Cisco experience will be quite comfortable with them.
As mentioned earlier, Ubuntu has support for load-balancing network traffic, using IPVS. Such configurations, however, tend to then turn the load balancing server into a single point of failure. To alleviate this issue, the keepalived package provides an implementation of the VRRP protocol for network redundancy. Rather than just having a single IPVS server, two or more servers are used, each with keepalived configured to ensure that one (and only one) of those servers has a network interface with a particular IP address. Should that server fail, at some point, one of the other servers will then assign the IP address to its own network interface, allowing continuity of service.
Redhat Cluster Suite
One feature of Ubuntu that isn't yet available in Debian is Redhat's Cluster Suite. The suite provides GFS, a local filesystem that can be mounted on several servers at one time; typically via a SCSI, SAS or fibrechannel disk array, but presumably using iSCSI, or ATA-over-ethernet also.
Unfortunately, specific documentation for running this on Ubuntu wasn't readily available. Installing the suite resulted in a number of errors, primarily due to the lack of a configuration file. Not really knowing where to turn, I first created a test installation on a couple of servers that I had running Redhat Enterprise, and then copied the configurations across to Ubuntu.
After a quite a bit of messing around, I eventually got it to the stage where all the packages were installed properly, and it started up correctly, and sure enough, I had a filesystem that was mounted on both servers. I feel that this is possibly a touch too convoluted for most people, however; it would be much nicer if a default configuration could be created that would at least allow the packages to install correctly in the first place.
Conclusion
In general, I found Ubuntu 6.06 Server to provide a good range of software for use on production machines. The various packages generally seemed to be in a much more ready-to-run state than I've found for comparable Linux Enterprise distributions; there's certainly much less messing around to get the system going.
All that remains to be seen now is whether companies such as Oracle will come to the table, and certify the system for use with their software. I'm sure such an event would then provide the answer to the wishes of many a system administrator; to have a homogenous network of only Debian-based servers, without the need for one or two Redhat or SuSE servers purely to handle Oracle.
If you found this article helpful, consider making a donation to offset the costs of running this server, to one of these addresses: