Patterson, Gibson & Katz : The Three Wise Men of RAID

While doing some research on the subject, I happened upon the original paper of David Patterson, Garth Gibson and Randy Katz on the potential for RAID devices with respect to performance increases.

It's a good read if you're interested in the origins of RAID and their methodologies.

A Case for Redundant Arrays of Inexpensive Disks

The "Inexpensive" comes from a time where you could buy either expensive harddrives that were reliable, or inexpensive harddrives which had errors now and then.

This came up with RAID, or mirroring, and later hamming-code and parity. In other words, they used redundancy to transform several "unreliable" disks into one "reliable" disk. If one disk had a bad sector, the other would have a copy.

Today, the average disk used in a RAID configuration, is more expensive than one used without RAID. RAID had the opportunity to increase performance, reliability, uptime and data security.

In the future, RAID won't be used as much in the traditional fashion. SSD drives are alot more reliable so that redundancy is not a must-have for home users and even companies (who have backups anyway). Also, using RAID to enhance performance of SSDs might be the wrong way, as they should implement this in the drive itself. Meaning, if an 80GB SSD has 8 channels (8 parallel operations) a 160GB disk might have 16 channels, doubling the bandwidth of the drive. Right now we don't see this as the 160GB-controller is the same as for the 80GB. But SSDs have all the potential to do this, and there is no limit on how much MB/s you can get. In theory, you could access all flash sectors at the same time, having nearly infinite speed set aside latency.

On the other hand, RAID might become an integral part of a filesystem, using redundancy from multiple disks to enhance reliability, in an era where storage capacity is booming this might suit users who don't need as much but just want it to work for 10+ years. ZFS is such a filesystem that has its own RAID engine, and it uses this to its advantage:
- it stores all its metadata on all disk members, so with 3 disks in a RAID0 config, you still have all the meta-data stored on all three disks
- it uses checksums to determine if a drive has "corrupt" data for a given file, and reads it off a redundant disk instead, then it fixes the corruption on the original disk. This way file corruption could be prevented; no application ever gets to see corrupt data from the disk.
- allows you to be flexible with the principles of RAID0 and RAID1. If you won't need redundancy, then don't. If you do, you can set that specific directory to store 2 copies of each file, so you would have a RAID-1 situation for just that folder.

So RAID lives on in the future, but in another form. The traditional RAID systems never reached its full (theoretical) potential due to smart parity RAID designs being scarse and the powerful host hardware was unutilized because everyone insisted on having Hardware RAID. Microsoft never paid true attention to RAID, where they could have set new standards if they released a WinFS with internal RAID, or conventional RAID on NTFS which is a little more sophisticated than what's available now. The RAID5-engine in WinXP is a joke, and in my own tests i already lost the entire array, with all (test) data on it. Don't want to store my files like that.

RAID-5 was all about reliability, right? But somehow bad drivers or hardware implementation bugs might cause bad things to happen. Rewriting in an improper disk sequence, attaching the wrong disk, not storing its cold/hot state, etc, could all cause total data annihalation. Either we need high-quality software drivers, or proven hardware designs. Currently for Windows the only option is the latter.

This is way before my time. My first harddrive was an 800MB WD Disk.
I believe still 5,25" format, don't know about the prices that time but it was casual hardware.

But its clear the priorities at the time RAID was "invented" or first implemented, are different than today. That doesn't mean RAID is wrong for purposes used today. We can see its usage and potential change very quickly if you mix RAID and a filesystem. However, the acronym is bad chosen, since RAID0 offer no redundancy, and JBOD/concat does not have a RAID-designation. Well if RAID0 is RAID, JBOD surely is too. Calling it a Disk Virtualization Engine would be more accurate today.