Headache with RAID options

[Antimatter]

I'll soon have 5 to 6 harddrive array that I want to setup RAID, so I can ensure that my data is safe even after a few harddrive failure, yet retain some performance and other stuff, anyway I know that RAID is not a replacement for backup, and will be backing up my more important stuff, but anyway at the moment I'm trying to figure out what kind of RAID layout I want, and I'm currently getting a massive headache trying to figure out a good layout.

Anyway here's the RAID options that are available to me:
--Each harddrive is separate
--JBOD/Raid 0
--RAID 1
--RAID 5
--RAID 6
--RAID 0+1
--RAID 10
--RAID 15
--RAID 50

My Analyzation:

Separate Disk, JBOD, RAID 0 --- These aren't really much of an option, esp with 5-6 harddrive, so I'm not going to really consider these options.

RAID 1 --- This could work with 6 harddrive, then i would have 750 gigabyte, and my data would be safe from a minimum of 1 harddrive to max of 3 harddrive loss, however one of my concern is it will only take one harddrive loss before it'll be in an degenerated state. Plus I'm not sure if I can do a RAID 1 of 6 harddrive?

RAID 5 --- This probably won't work, because the loss of one disk will put it into degenerated state, plus the whole issue with party overhead, and silent harddrive corruption, corrupting other stuff. Because we know that harddrive don't always right away fail, there might be data corrupt and this probably won't protect verses it.

RAID 6 --- An option, but we got additional overhead of double party calculation, but its currently one of the only real RAID offering that can guarantee the loss of 2 disk minimum before the array enter fully degenerated mode, but we got the issue of party overhead and the issue of RAID 5 also

RAID 0+1 --- A possibility, but i think RAID 10 would be better in this case, it can suffer from 1 to 3 harddrive lossage, depending on where. This one would certainly be fast without having to worry about party calculation

RAID 10 --- Once again akin RAID 0+1 i can loss from 1 to 3 harddrive....

RAID 15 --- Can lose one to three harddrive before entering degenerated mode, but we got the party calculation issue and other issue of RAID 5

RAID 50 --- Can lose one to two harddrive, but would be in degenerated to seriously degenerated mode and again got party/calculation issue of RAID 5


Data requirement:

The primary purpose of this array is to prevent time consuming restoring, and to assist with proving me with uptime, if an harddrive fails, the odds is I'll keep the unit shut down until a replacement harddrive has arrived. But its not guaranteed unfortunately.

It'll also be nice if it can sustain at least a single drive speed of read/write, while being able to do it with my CPU utilization at maximum if possible.

The harddrive array will be mainly for running VMware images, editing very large files, temporary swap space for programs that requires vast amount of memory, allots of compiling, and probably some other light webpage/database duties.

The main thing I'll be storing on it probably will be some movies, some photos, and several VMware images, and other crap. The most important stuff like document, and configuration/setting and important pictures will be backed up to DVD for safety.


Anyway I'm wondering what your guys/gals recommendation is for me for an good array setup, I'll admit I am partial to RAID 6, because i can lose up to 2 harddrive guaranteed before It'll become fully degenerated, all of the other array can only loss an guaranteed one harddrive before being in a degenerated mode. However the negative of RAID 6 is its deviated from RAID 5 with all of the flaws and problems of RAID 5.

So anyone got any good recommendation here?

 

[dirtydr]

As you already know one array will not fit all your needs so why not run a 2 drive level 0 for the more I/O intensive side of things (swapfile/editing/etc) and a 3 drive level 5 w/hot spare for the more sensitive data. Not sure which controller you're running but it should support multiple arrays. Just an idea . Whatever you end up going with, its always nice to have a spare drive laying on the shelf to prevent the replacement scenario you have already explored.

"It'll also be nice if it can sustain at least a single drive speed of read/write, while being able to do it with my CPU utilization at maximum if possible."

Did you mean minimum CPU utilization here? If so, the key is in picking a card that has a dedicated XOR processor which means a cheepy isnt going to do it. Have you already picked out the controller?

 

[uOpt]

What you want is 0+1 or 1+0 with more than two units in raid-1.

With 6 drives I would go 2x raid-0 by 3x raid-1, resulting in net capacity of 2 * raw_harddrive.

In general, raid 0+1 and 1+0 even with just two units in raid-1 are more robust than people think since an array degraded from one disk failure will only be killed by a second drive failure if it hits the matching disk. (Only if the software is smart enough).

The only alternative is raid-6 but you won't like the write speed.

 

[prlwytkovsky]

For us only raid 6 works. We store lots of patient data, which doesn't get accessed regularly. This is a problem with raid 5 since a disk can develop bad sectors and we would not notice. Then an other disk would fail and we lose the redundancy to restore the bad sectors on the other disk... Patient data lost So now we have only raid 6. We still have to run regular verifies on the raid but we can lose two disks now without losing data. The Areca cards are nice, they signal a faulty disk, you just swap it out with a new one and the raid array starts rebuilding automatically.

With six disks in a raid 1 (mirror) you cannot loose three disks. The chances that one of the faulty disks is not a mirror of another faulty disk are minute. You need to have true redundancy like in raid 6. In raid 6 you can loose any two disks of the six.

 

[Antimatter]

[double post] : stupid browser

 

[Antimatter]

I'm not holding any truely critical data like prlwytkovsky is, and I will admit that I like the idea of RAID 6, but the issue is the write speed, because this disk will probably be accessed quite often, afterall it'll be holding vmware images, and various other files that might get changed.

Anyway I know I have several types of data that will be stored on the arrays.

Type 1
--Important to Critical document
--Personal source code
--Personal photo/movies
--Other

Size: What I currently have in this catalogy would fit aproximatly 2-4 dvd worth of media, and this catalogy probably will be almost religiouly backed up within a few weeks. Anyway its mostly many small files, and many large files as in megabytes in size, rarely have anything larger than a dozens of megabytes here.

Speed: Not absolutly critical to have high speed access, but should not be abyssal. Write speed is nice to have, but i would suspect read would be more important.


Type 2
--Downloaded programs
--Downloaded movies/picures etc
--Other important but not critical stuff that can be replaced

Size: Varriable, probably 10-50+ dvd if all backed up

Speed: Would be nice to have decent speed, but read speed is probably most imporant in this sitution, this will be mainly used as a storage area to hold this kind of stuff.


This is probably some of my largest problem, these are data and stuff that I would really like to keep safe, but I just don't have enough time or space to make backup, and if i lose it, i'll be upset, and will have to spend alots of time recovering it, but its not irreplaceable.


Type 3:
--Large downloaded stuff, usually via torrent, or from friends or what so not
--Can be replaced fairly easly, but just extremely time consuming, due to a slow and crappy internet link
--Vmware images for virtual systems and so forth
--System files/settings and so forth

Size: Probably a few gigabytes for system files, to tens to hundreds of gigabytes for vmware images, downloaded files, and extremely large archives/files

Speed: Fairly important to have atleast disk speed minium if not higher at all possiable, but RAID 0 is not desired, because of the time it will require to rebuild the system, which is in tens to hundreds of hours.

This is the area that holds the main system stuff, and large crap, now I'm not quite sure if i should put system in type 2 or type 3, because setting up the system can be quite time consuming, but its not critical, I suppose I can try my best by trying to have regular backups of the system so I can rapidly restore the system after a disk array crash.

Type 4:
--Swap disk
--Temponary files
--Torrent downloads
--Downloading, and other stuff
--Heavy reading/writting temponary files kind of stuff

Size: Can be very very large, up to gigabytes worth of data, mostly temponary kind of files, or files that I don't care if they are lost.

Speed: Very important to have atleast disk speed in read/write, preferly as high as possiable, for swap disk, temponary files and so forth


Conclusion and other remarks

Now i hope that more detailed information above should assist you guys in helping me figure out what a good RAID layout will be for my system, the odds is it probably will be moderately to decently complex layout because I have various type of requirement and type of data.

System configation
6x 250 gigabyte harddrive.

However I have't ordered the system yet, and if needed I can order the rest of the system but defer the ordering of the harddrives theirselves. I'm beguning to wonder if 6x250 gig isn't the best choice, I might be better off with some harddrives that are more focused on high speed access such as raptors, and some disks that are focused on reliability and storage, so perhaps a harddrive to a few harddrive of raptors for high speed temponary storage, then a few Seagate enteprise drive for high reliable storage?

[edited] updated to add in more detailed information.

 

[defakto]

With six disks in a raid 1 (mirror) you cannot loose three disks. The chances that one of the faulty disks is not a mirror of another faulty disk are minute. You need to have true redundancy like in raid 6. In raid 6 you can loose any two disks of the six.

I'm not sure where you're getting this info from. RAID 1 mirrors the data across drives as it is written, it doesn't mirror the physical sectors of the drives. If the data is corrupt when written you would have a problem, but this would affect all raid levels, 6 included. Once it's actually written to disk, you should be ok, in fact good drivers should notice the inconsistency and report an error.

Maybe I'm misreading what you're writing.

 

[Antimatter]

I'm not sure where you're getting this info from. RAID 1 mirrors the data across drives as it is written, it doesn't mirror the physical sectors of the drives. If the data is corrupt when written you would have a problem, but this would affect all raid levels, 6 included. Once it's actually written to disk, you should be ok, in fact good drivers should notice the inconsistency and report an error.

Maybe I'm misreading what you're writing.

What I mean by that, is for example let's say we are in a RAID 5 array that is in degerated mode, and we are rebuilding the array, now if a sector is corrupted, that data will result into a corrupted party check and possiably propogate thoughout the array, I'm not quite sure, if needed i can dig up the link and post it here.

 

[defakto]

It still wouldn't matter, the parity check on the data should fail then at some point if the data is corrupted after it's written to the disc. That is a large portion of why parity exsists. If the data is corrupted on the disk and the parity check reflects the corrupted data then nothing would fix it, no amount of raid level would help.