The problem is that we have a chicken and egg problem. Very few programs (including most enterprise databases) use AIO. Why? For portability reasons; there

The problem is that we have a chicken and egg problem. Very few programs (including most enterprise databases) use AIO. Why? For portability reasons; there are other ways of doing things (i.e., using thread pools) that will allow a database developer to get all or most of the benefits of AIO, at least before the days of super fast storage. (Now that we have really fast PCIe-attached flash which is fast enough that scheduler overhead starts becoming a real problem, this may no longer be true. As a result, there is little incentive to improve AIO (on all systems, but especially on Linux --- a lot of the Direct I/O work was done to make the enterprise database vendors happy). And since AIO isn't good enough, very few people want to use it, and since making it better is difficult, few people are interested in working to solve the problem, and the cycle repeats again.