How RAID 5 lays out data
RAID 5 stripes data across every drive and adds one parity block to each stripe. The parity block moves from drive to drive, so no single disk takes all the parity writes. It needs at least three drives.
The price is one drive's worth of space, however many drives you use. Four 4 TB drives give 12 TB usable from 16 TB raw, or 10.9 TiB when you count in binary units. Parity takes 33% of the space with 3 drives, 25% with 4 and 13% with 8. So the loss shrinks as you fill more bays, but a bigger array also has more drives to read when one fails.
What happens when a drive fails
RAID 5 survives one failed drive. The array keeps running in a degraded state, and each read that would have hit the missing drive is worked out from the others, which slows it down. When you install a replacement, the controller reads every block on the surviving drives and writes the rebuilt data to the new one.
That read is the exposure. On four 8 TB drives it covers 24 TB. At the RAID Calculator's modeled hard drive speed, one 8 TB drive takes at best 14 hours. A busy array takes longer. A second drive failing in that window loses the array.
A read error can do damage too. At a spec rate of 1 in 10^14 bits, the chance of at least one unrecoverable read error across those 24 TB is at most about 85%, and at 1 in 10^15 bits it's at most about 17%. Those are ceilings, not forecasts. But RAID 5 has no spare parity left to cover a bad sector in the middle of a rebuild.
What RAID 5 does to writes
Reads are quick because every drive serves data. Writes are where RAID 5 pays. Changing a small piece of a stripe means reading the old data and the old parity, working out the new parity, then writing the new data and the new parity. That's two reads and two writes for one small write. A write that fills a whole stripe skips the reads, because the controller already has all the data it needs to calculate parity.
So large sequential writes are fine, and lots of small random ones, like a database or a stack of virtual machines, are not.
There's a second hazard on power loss. If the machine dies between writing the data and writing the parity, the stripe is left inconsistent, a gap called the write hole. Hardware controllers cover it with a nonvolatile write cache, Linux md can use a journal, and ZFS's RAID-Z1, its single parity mode, avoids it by giving each block its own stripe width.
When RAID 5 is the right pick
RAID 5 suits a read-heavy array of three to five drives, like a media library or a backup target. It keeps the most space of the redundant levels. Four 8 TB drives give 24 TB in RAID 5. The same drives give 16 TB in RAID 6 and 16 TB in RAID 10.
As drives grow, the rebuild gets long enough that a second failure or a read error stops being a remote chance. RAID 6 gives up one more drive of space to cover that.
Whichever level you pick, RAID guards against a dead drive. It doesn't guard against a deleted folder, ransomware or a power supply that fails and damages everything at once. Keep a copy somewhere else.