Use case RAID5 on LVM mixed HD sizes

I can't read German and their is only so much text you can put into Google translate. Is there an English link somewhere.

If its only a speed hit then I'm OK with that. With 4 drives the math just works. All the rack mount cases use 4 disks per row. Each row a raid 5 or 2 rows for a raid 10 setup.

I'll summarize it up for you later or tomorrow! (in english)

Greetings
David

So. Translation done...

Zitat

Impact on the Write-Performance

In difference to the read performance is the measurement of the write performance on raid5 way more complicated and depends on the amount of data that has to be written and the count of the hard disks.
Assuming that hard disks with less than 2TiB have a atomar block size (also called sectorsize) of 512Kb.
Lets assume that we have a raid5 with 5 disks (4/5 data and 1/5 parity) the following scenario takes place: If an application wants to writ 2048Kb of data it will write 512Kb to each data disk in the best case scenario, whil (of course) one block doesnt hold userdata (but the parity). [...]
If an application now wants to write a single block of 512byte, a bad case happens. It needs to read the block that has to be change including the paritiy block. It then has to recalculate the paritiy and after that the two 512byte blocks can be written. This means double effort to read and double the effort to write a block.
If you would say that reading and writing takes the same time, the effiency, the so-called RAID 5 write penalty, would still be 25%.
In practice this wordt case scenario mostly would not occur to a RAID 5 with 5 disks, because filesystems often have blocksizes of 2kB, 4kB or more and show practically only the well-case-writebehaviour.
The same applies to a RAID5 with 3 disks. However the behavior of a RAID 5 with four disks (3/4 Data and 1/4 Parity) differs. If the system now wants to write a Block of 2048 Byte, it has to do two writes. 1536 Byte with Well-Case-Performance andagain 512 Byte with worst-case-behaviour.
To reduce this worst-case-behaviour there are cache-strategies, but is still implied (on a RAID 5) to use a set of two, four or maybe even eight disks for userdata plus one disk for the parity. Because of that RAID-5-systems with 3, 5 or 9 disks do have a particularly favorable performancebehaviour.

Alles anzeigen

Source is still the german wikipedia article for RAID.

Greetings
David

I can live with that. It seems that a 2048 KB file system is much better than 512 KB file system. Like I said before 5 disks the math does not work and harder to keep track of which disks are in which array. A 4 disk system each row has an array. I do mostly read cause its a storage has. I'm sticking with 4 disks arrays and LVM management. I'm not to worried about write performance but I'm sure a 5 disk array would be faster than a 4 disk array. Someone else had told me this a long time ago he had said that the 9 disk array was fastest of them all.

Well... I can only tell that my 5 disks RAID5 (and I build it before I knew of this best/worst case scenarios) is running with so much speed, that I guess that only the 9 disk array could top that speed, but even with 5 drives, the SATAII lanes are nearly maxed out. Don't know how much speed the controller can saturate through the PCI-E lane or if it is limited @600MB/s like SATAII is...

Greetings
David

In reality I only need 200 megs per second anything more than that is gravy. I am planning on bonding 2 gigabit NICs hence the need for 200 megs. I would use 5 disks but 4 disk arrays are just much easier to use when using rack mount cases. So if I get 500 megs with the hardware raid per array I'm happy more than enough I will ever need. That box would easy handle about 8 users cause that is the max I'll ever have at the same time.

Edit/Update: Nevermind what I posted in my previous post... SATAII is limited @300MB/s, so I'm already above that.

Greetings
David