Really. A month ago I turned on AHCI on a drive, and after bouncing the box, I was sent to single user mode.
Yea. I considered that, and it looked like it would likely be a long, and arduous process. For an unknown reward, if any.
That was exactly my point. Performance tuning is hard work. On a file system on a single disk drive for common desktop/development workloads, there is little performance benefit. The amount of work is large, and the risk is high. In the case of turning AHCI on, you experienced one particular risk.
Another risk comes from the fact that some "tuning" operations may require moving the data off and back on, which for the average home user (with just a few drives) is tedious, and has the risk of a mistake wiping everything out.
Now, if this was a large professional-strength file system product, where tunables could be changed online, which could handle multiple drives, supported online migration, and where there was considerable industry experience in tuning, then it would make sense. If you are a large bank, supercomputer center, or secretive government agency, and have spend tens or hundreds of millions on your storage and file system, you will be very willing to spend a few hundred thousand on consultants to tune it, and get 10% or 20% better performance out of it. For the single-disk desktop user, default settings tend to do pretty well.
I have a couple, and am looking forward to replacing everything with SSD's. I'm only waiting, because I'd like to get some more numbers. Where longevity, and speed, are concerned.
I have two SSDs in my home system, I boot from one, and use the other for backups, or an alternate boot during upgrades. The speed difference to hard disk is astonishing. The way I like to put it: In engineering, 10% improvement is a big deal; in astrophysics, you look for factors of 10. In that sense, SSDs are like astrophysics: they are an order of magnitude faster. And that's using a cheap consumer-grade SSD (an Intel device with a SATA port); the high-end enterprise SSDs are even more astonishingly fast.
BUT: You have to be careful with write endurance. The high-capacity SSDs that are shipping now are somtimes spec'ed to be overwritten on average once per day, and then are expected to give you a useful life (on average! no guarantees!) of 5 years. Since their interfaces are typically capable of delivering about 500 MByte/s, or 43 TB per day, if you are careless, you can overwrite say a 1TB SSD about 43 times a day, and just that simple back-of-the envelope calculation reduces their expected life span to a little over a month. In this is even before write amplification, which in the case of random 512 byte (sector) writes can be another huge factor. (In reality, many drive firmwares will balk, refuse the writes by reducing their speed, and maintain reasonable lifespan, and many modern SSDs expose 4K sectors). So when planning your file system layout, if you have a write-intensive part, it might be a good idea to segregate that onto a separate file system that uses either a spinning disk, or RAIDed and disposable SSDs, if you can afford that. I thought about it, and decided that the only write-intensive part of my root disk is /var/, and the total write traffic to it is a few MB per day, so on a multi-GB SSD the overall write rate is very little. But for other workloads (DVR for example), the answer might be radically different.
Also to consider: In terms of raw bandwidth (in units of MByte/s per $), spinning rust hard disks are still impossible to beat at large capacities, if your workload is suitable for the head/platter interface (large IOs and long sequential streams). So if you have something bandwidth intensive, it might be cheaper to buy a half-dozen discounted traditional disks and stripe them together, rather than one expensive SSDs. But beware of the overall reliability when you do that.
Again, all this really depends on the workload. For a normal desktop system (with e-mail, web, some documents, a few baby pictures being stored, and some development work), a single spinning disk with a stock file system is usually sufficient, and a good compromise between cost, convenience, and performance.
