The answer to this question is, of course, no one. But it does bring a few laughs from an audience.
But the interesting follow-on question is this one:
Why aren't all systems High-Availability?
The answer to this question has two parts:
This is a truly interesting question - and the fact that it isn't often asked, and less often answered is even more interesting. But before discussing why this might be, let's explain how cost and complexity are the blocking issues in most people's minds.
Traditionally, HA systems are very expensive. Expensive and specialized hardware running on expensive proprietary systems was often required. Expensive software (often costing hundreds of thousands of dollars) was required. This meant that HA was reserved for a very few choice applications, where the enormous cost could be justified. For the Traditional HA world, this meant that cost was such a large objection, that complexity didn't even enter the picture in the minds of most people.
However, with commodity hardware running open source software (like Linux, Linux-HA and DRBD[1]), the cost factor almost completely vanishes. HA systems with no SPOF[2]s can be assembled for a few thousand dollars total cost (and sometimes even for less). Replication techniques like DRBD[1] are an essential part of this picture.
With the cost barrier almost completely eliminated, a new factor shows up as the major limiting factor for HA acceptance - complexity. This is because no matter what you do, two machines is inevitably going to be at least a little more complex to manage than one. This is another reason why we hate complexity[3], and continually strive to keep it under control.
One of the conclusions that you can draw about so few people considering HA seriously for most of their servers is that people don't even consider using HA in most applications - and in many cases aren't even aware that they aren't considering it. They take it for granted that they can't have it.
AlanRobertson[4] believes that as OSS HA software begins to become well-understood and well-accepted, that the inevitable result will be a huge growth in HA systems. This is similar to what happened with the PC. Before PCs, computers were too expensive to use in any but the most critical and demanding of tasks. But, as the vastly more cost-effective PCs came out, people began using them for applications which no one would have ever dreamed of using mainframes for in the past. Today, the majority of uses PCs are put to would have been literally unimaginable 30 years ago. As acceptance of OSS HA software grows, it will be the same for HA systems in the future. Many of the uses people will put HA systems to in the future will likely be unimaginable today.
This trend is already underway today, with OSS HA systems being used for commodity firewalls, badge reading systems, proxy caching servers and many other commodity applications.
OSS projects get their developers from their users. A certain percentage of users of open source software become developers for it. With this vast expansion of users, will inevitably come a vast expansion of developers. This in turn, will bring more features, which bring more users, which will bring more developers...
| [1] | http://www.linux-ha.org/DRBD |
| [2] | http://www.linux-ha.org/SPOF |
| [3] | http://www.linux-ha.org/ComplexityAndReliability |
| [4] | http://www.linux-ha.org/AlanRobertson |
| [5] | http://www.linux-ha.org/ThePyramidPicture |
This information provided courtesy of the Linux-HA project at http://linux-ha.org/