Every RAID level is a trade between space, speed and safety - and the trade only becomes visible on the day a drive dies. Here is each level with its real numbers and the failure story that teaches what it actually does.
Vendors sell RAID levels by their best feature. Engineers judge them by their worst day. This page gives you both for the five levels a small-office server will ever meet - what you get, what it costs, and what happens when a drive fails, including the RAID 5 rebuild story that has emptied more than one office's data. Worked through with four 4 TB drives so the capacity maths stays comparable. And the standing reminder for the whole set: whichever level you choose, RAID is not a backup - it covers drive failure, nothing else.
RAID 0 stripes data across every drive, so reads and writes spread over all of them and performance is excellent. It is also the only "RAID" level with no redundancy at all - the R stands for redundant, and RAID 0 is the exception the industry politely ignores. If any single drive in the set dies, the array is gone, because every file is spread in pieces across all the drives. The failure story is always the same: a scratch disk or a video-editing cache built on RAID 0 for speed, one drive clicking on a Monday, and a weekend of work unrecoverable. Legitimate uses exist - temporary rendering space, rebuildable caches, anything you can recreate in an hour. Server data is not one of them.
Usable space: n x drive size - 4 x 4 TB = 16 TBFailures survived: 0RAID 1 writes the same data to two drives at once. You get half the space you paid for, and in return the failure story is boring in the best way: one drive dies, the server keeps running on its twin, you swap the dead drive, the mirror rebuilds - a small array rebuilds in a short time because it is a straight copy, no parity maths. Performance is simple and predictable, and any controller (or Windows itself) can do it. This is why RAID 1 is the standard home for a server's operating-system volume, usually on a pair of SSDs: the OS does not need much space, and boring is exactly what you want from it. Its limit is scale - mirroring is a two-drive answer, and using it for a large data volume wastes half of a lot of terabytes.
Usable space: 2 x 4 TB = 4 TB usableFailures survived: 1 of the 2RAID 5 spreads data plus parity - a kind of checksum - across three or more drives, so any one drive can die and its contents can be rebuilt from the others. It is space-efficient, which made it the default office choice for years. Two honest problems have grown around it. First, writes pay a penalty: every write means reading the old data and old parity, then writing new data and new parity - four operations - so RAID 5 writes are slow, which databases feel. Second, and worse, is the rebuild. Modern drives are huge, so rebuilding a failed 8 or 12 TB member means reading every remaining drive end to end, for many hours or days, under full load. Desktop-class drives carry an unrecoverable read error rating around 1 in 10^14 bits - roughly one error per 12 TB read - so on a big array the maths says an unrecoverable error during a rebuild is a realistic event, and a second failure or an unreadable sector at that moment loses data. That is not theory; it is the standard cautionary tale of the last decade. RAID 5 on small SSDs is fine. RAID 5 on large spinning drives is a risk you should take knowingly, not by habit.
Usable space: (n - 1) x drive size - 4 x 4 TB = 12 TBFailures survived: 1 - and the array is naked until the rebuild finishesRAID 6 is RAID 5 with a second, independent parity block. It costs one more drive's worth of space and a heavier write penalty - six operations per write - and in exchange it survives any two drive failures. That single change transforms the rebuild problem from the previous section: while a large drive rebuilds, you still have a full layer of protection, so an unreadable sector or a second tired drive is an inconvenience rather than a loss. For arrays of big HDDs - the 4, 8, 12 TB class - RAID 6 is the honest default. The trade is real: write-heavy database workloads feel the double-parity cost, which is why the next level exists. But for file storage, archives, backup targets and general office data on large drives, calm beats clever.
Usable space: (n - 2) x drive size - 4 x 4 TB = 8 TBFailures survived: any 2RAID 10 builds pairs of mirrors, then stripes across the pairs. You give up half the raw space - the same price as RAID 1 - and in return you get the best combination most offices can buy: reads and writes are fast because there is no parity maths on the write path, and a rebuild is a simple mirror copy from the surviving twin, which is quick and puts far less strain on the remaining drives than a parity rebuild. A database - SQL Server, or the data files behind Tally and similar packages - under real multi-user load is exactly the workload that punishes RAID 5 and rewards RAID 10. Virtual machines too: several VMs sharing one array all write at once. One nuance to state honestly: RAID 10 survives one failure per mirror pair, so two failures in the same pair still lose the array - unlikely, not impossible. With hot spares and prompt swaps, it remains the level engineers reach for when the data works for a living.
Usable space: n x drive size / 2 - 4 x 4 TB = 8 TBFailures survived: 1 per mirror pairThe pattern behind all five levels: match the trade to the data. Operating system on a mirrored SSD pair. Databases and VM storage on RAID 10. Large-capacity file storage on RAID 6. RAID 5 only where drives are small or SSD-based and its rebuild window stays short. RAID 0 only for data you can recreate before lunch. And then, whatever you picked, arrange the separate backup - because every story on this page is about drives failing, and drives are the easy disaster. The hard ones are on the next page of this set.
RAID levels and server hardware explained live - capacity, failure stories and the right level for each job - in the full video.
Watch Full VideoBecause of the rebuild. When one drive in a RAID 5 array fails, rebuilding its replacement requires reading every remaining drive completely. With multi-terabyte drives that takes many hours or days, and the odds of hitting an unreadable sector - or a second failure - during that window are real, especially with desktop-rated drives. RAID 6 survives that second event; RAID 5 does not.
Not always - it costs half your raw capacity, against one drive in RAID 5. RAID 10 wins on write speed and rebuild safety, which is why databases and virtual machines belong there. For a large archive of files that is mostly read, RAID 6 usually gives the better balance of space and safety.
Usually not in any way you should plan around. Some controllers offer online level migration for specific paths, but it is slow, risky, and not available everywhere. Treat the level as a decision you make before the install, with a backup taken before any change - which is the whole point of planning it on paper first.