↝ OWC / MacSales.com... ↜
↝ diglloyd Deal Finder... ↜
Buy other stuff at Amazon.com...
HGST Ultrastar He8 Performance: RAID-0 vs RAID-4 / RAID-5 Compared 3/4/5/6/7/8 Drives
RAID-4 and RAID-5 are “striping with parity”.
A RAID-5 uses the equivalent of one drive capacity of N drives for parity information. It can be thought of as a RAID-0 stripe with one parity drive (“striping with parity”), for fault tolerance. RAID-5 uses distributed parity and RAID-4 uses a dedicated parity drive; functionality is equivalent, so the only question is whether one performs better than the other for any particular (specific) implementation.
Ideally, performance should scale linearly with the number of drives. Also, a 3-drive RAID-5 / RAID-4 should perform reasonably close to a 2-drive RAID-0 stripe.
$1649 SAVE $150 = 8.0% Apple 13.3" MacBook Pro with Touch Bar (Late 2016, Space Gray) in All Other Categories
$1400 SAVE $149 = 9.0% NEC PA272W-BK-SV 27" 16:9 IPS Monitor with SpectraViewII in Computers: Displays
$899 SAVE $200 = 18.0% Olympus OM-D E-M5 Mark II Mirrorless (Body, Silver) in Cameras: Mirrorless
RAID-4 vs RAID-5 vs RAID-0 sustained transfer speed
DiskTester run-sequential used to test peak speed of empty volume, freshly erased. Throughput figures are real-world figures through the OS X file system.
This is excellent: the RAID-4 and RAID-5 configurations with three drives are nearly as fast as a 2-drive RAID-0 stripe. That’s all one can hope for. However, the performance lags that of an equivalent RAID-0 stripe in general; the 5-drive RAID-4 still is not as fast as a 4-drive RAID-0 stripe. If maximum performance is required, striping is still the best option. As well, software RAID-4/RAID-5 with 4 or 5 drives does consume the better part of one CPU core, so it is not “free” in CPU terms.
Note that at least with up to 7 drives, RAID-4/5 are about as fast as a RAID-0 stripe with one less drive—excellent.
RAID-4 outperforms RAID-5 for reads using 4 or 5 drives, and is slightly less fast for writes. RAID-4 thus seems to be the better choice. However, the advantage is scarcely larger than the margin of test error (normal variation between iterations).