Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

Distributed Parity

The workaround for the increased wear is to move from a dedicated parity drive to a rotating scheme, where every drive will keep data and parity information. The only requirement is that the parity information and the actual data of any given file are distributed in a mutually exclusive scheme across the drives. That is, if data of file a are written to drive 1,2 and 3, the parity will have to be stored on drive 4. Likewise, if data are written to drives 2,3 and 4, the parity information will have to be stored on drive 1 and so on.

Level5 RAID

This distributed parity scheme in combination with block-level striping is called Level5 and is currently the most commonly used RAID level. Because the data are written in blocks rather than in bits or bytes, Level5 is substantially faster than Level3 with a better overall reliability of the array than what Level4 has to offer. However, keep in mind that the parity calculations are extremely computationally intensive, which is the reason to add a dedicated RAID XOR processor to the RAID controller in most cases. One would assume that modern processors should have no problem doing the parity calculations on the fly, however, this is not necessarily true.

In Level5 RAID, the parity (red blocks) are distributed across all drives within the array, which results in a more distributed wear of the drives as well.

Instead of using the CPU-usage numbers from whatever HDD benchmarks that are available and that are notorious for generating inconsistent numbers, we used the CPU-usage tab of Windows TaskManager during a run of ATTO (direct I/O) using a HighPoint RocketRaid 1640 "Soft RAID 0,1,5" controller in RAID Level5 array using four Fujitsu MHT2060AH SA - 2.5" native SATA drives. The "Read performance" in this case shows some 70 MB/sec sequential transfers, however, the write performance is capped at about 13.5 MBsec. These results are typical for Level5 RAID arrays. Note that the CPU usage during the entire execution of the benchmark is at 100% (P4 2.8GHz on Shuttle SB52G2, 2x1GB Mushkin DDR 400 (unbuffered) System memory).

Level6

Level5 or any of the aforementioned levels are using a single drive for the storage of parity data, the result is fault tolerance for single drive failure only. In case the fault tolerance has to be extended to more than one drive, the number of parity drives has to be extended as well, for example, Level6 RAID is using dual distributed parity, which can overcome failure of two drives in the array without data loss.

next page:    => RAID Cache and Journaling Memory =>