Things that make me go "Hmmmm"

[Rich A]

I often see people obsessing about drive failure with a Raid 0 array. I don't get it. If you have two 200 GB drives in a RAID 0 to give you 400 GB total and one drive fails .. you've lost everything! That's true

So what happens when you have ONE 400 GB drive and it fails? Don't you still loose everything? What exactly is the difference? Also please don't tell me about more drives meaning more chances of failure. That kind of thinking is really like the gamblers who say because a coin comes up heads 50 times in a row, the chances of it coming up tails on the next flip is greater. Sorry .. it is and always will be 50/50. And no matter how many drives you have, the MTBF rate is always the same.

Hmmmmm.

[StephaneM]

Originally Posted by Rich A

I often see people obsessing about drive failure with a Raid 0 array. I don't get it. If you have two 200 GB drives in a RAID 0 to give you 400 GB total and one drive fails .. you've lost everything!

Well, if you have two 200 GB drives not in a spanned drive config, if you loose one, you don't loose the other. If you span a logical drive on both you loose everything.

The point is that the spanned volume should be more efficient than a volume on a single drive (or two drive with non spanned volume). That's the difference with two 200 GB drive and 400 GB drive : performance. Spanned, you split load among the drives, non spanned you split load logically (page file, data file etc...)

The best setup is three 200 GB drives in RAID 5 : if you loose a drive you don't loose anything and you also split the load between the drives. The only disadvantage is that you loose 200 GB of capacity.

I think it's the best choice for HTPC if you don't want to loose any data because you can't really backup the kind of data an HTPC produce (because backuping files over 500MB is a long process - as well as restoring them-, those files cannot be compressed during backup and in the end you have to have a 200GB drive/device to back them up).

Regards,
Stéphane.

[queonda]

Originally Posted by StephaneM

The best setup is three 200 GB drives in RAID 5 : if you loose a drive you don't loose anything and you also split the load between the drives. The only disadvantage is that you loose 200 GB of capacity.

What happens if the one drive starts to fail (buy kinda working) and starts corupting your files? When you replace the drive and Raid starts rebuilding from corupt files. Is this possible?

[StephaneM]

Originally Posted by queonda

What happens if the one drive starts to fail (buy kinda working) and starts corupting your files? When you replace the drive and Raid starts rebuilding from corupt files. Is this possible?

Usually in a RAID5 config when a drive is failing then it is really failing => meaning that the RAID controller discard the drive totally and use the remaining ones.

Normally the RAID controller is supposed to "suspect" a failure when the parity doesn't match the data (data block by data block) and consequently will try to repair the error if it cannot => it mark the drive with the wrong data as failed.

When one of your a drive is in a failure state you can either :

* shutdown the machine, remove and reconnect the drive and see if the situation can repair itself (it's often the case with good SCSI hot plug drive => oxydation on connectors can cause parity failuer)

* shutdown the machine and replace the bad drive by a new identical one (same capicity or larger, same speed etc..)

When you restart your system you just have to assign the new drive in the logical volume and let it rebuild the drive (depending on how large is your drive this can take up several hours but the machine is usable during the rebuild time : the only recommandation is that the machine should not crash or get a power failure during rebuild because with some (cheap) RAID controllers this could lead to a total logical drive failure (but you can usually recover it).

So the bottom line is : you cannot have corrupted files in a RAID5 setup because of one drive failure. Only software failure can cause corrupted files (exactly the same way with non RAID drives).

Regards,
Stéphane.

[queonda]

Thanks StephaneM

Sounds like you've been messing around with Raid for a while (i don't mean the bug spray).
I haven't done a raid setup yet, but that's my 2006 New years resolution. I've heard of 1+0 is a good raid. What's your opinion?

[StephaneM]

Originally Posted by queonda

Sounds like you've been messing around with Raid for a while (i don't mean the bug spray)

I use them for work : RAID5 for database server, RAID 1 for other servers.

I've heard of 1+0 is a good raid. What's your opinion?

It's depend on your budget and what you want to achieve, so I'll begin with a few facts :

RAID0 = you distribute your logical volume betwen drives, with no redudancy (loose one drive, you loose everything). This is used for maximum write/read performance. Minimum two drives.

RAID1 = Mirroring, you copy drive content onto another one. Redundancy is great. Performance are good for reading, fair for writing. You loose 50% of drive capacity (two drives)

RAID5 = Distribute logical volume between at least three drives with parity. A block of data is written on drives like this : Part A on drive A, Par B on drive B, Parity on drive C. A and B are the useful data, C only helps reconstruct A or B with the remaining one. A,B,C are not always on the same drive for better peformance. Redundancy is great. Performance are very good for reading, fair for writing (poor reading / writing when rebuilding). You loose 30% of drive capacity with three drives (25% with four, 20% with five and so on)

RAID 0+1 or 1+0 = you combine a mirrored volume with a spanned volume. Either you span a volume on two mirrored volumes, or you mirror two spanned volumes. Performance are good for reading and writing. Redudancy is great. You are loosing 50% of drive capacity (four drives)

So RAID 0+1/1+0 achieve the best performance for a redudant solution, but for a higher cost than RAID5.

I personally choose RAID5 because

* I do find performance to be enough for HTPC
* It's cheaper
* Only three drives needed, HTPC cases are generaly limited in space and also it is always recommended to have a dedicated power line (from the PSU) to each drive: for redudancy reason and also because two drives on the same power line can issue a failure of one of the drive (power drain).

Regards,
Stéphane.

[Eric3a]

Sorry .. it is and always will be 50/50. And no matter how many drives you have, the MTBF rate is always the same.

The stats seem wrong to me.
Because the "M" in MTBF is for "Mean".
No disk fails at it's MTBF. But some fail before, some after.
So by putting 2 drives working together the combined drive will fail at the SHORTEST life of the 2 drives.

Push the reasonning further: If you had an infinite number of drives, and the failure of anyone of them took the whole thing down, you know that the unit wouldn't work because you know statistically at least one drive would be dead on arrival and just wouldn't work.

Some rare disks fail way before their MTBF, others will last a lot longer.
But as you add disks, the actual failure time of the unit becomes that of the shortest life drive.
Which if you're unlucky like my latest raid disk was within 1 hour of receipt.

Perso, for videos I don't care...


But the stats just didn't seem right. Did i get it right?
Eric

[merrypig]

You are correct Eric.

The MTBF of the _system_ reduces but the MTBF of any given drive is unchanged. (*)

(*) actually it will change slightly since some hardware/cabling/psu factors are often common so added a 2nd disk can impact the first anyway...

[Rich A]

Originally Posted by StephaneM

Well, if you have two 200 GB drives not in a spanned drive config, if you loose one, you don't loose the other. If you span a logical drive on both you loose everything.

The point is that the spanned volume should be more efficient than a volume on a single drive (or two drive with non spanned volume). That's the difference with two 200 GB drive and 400 GB drive : performance. Spanned, you split load among the drives, non spanned you split load logically (page file, data file etc...)

The best setup is three 200 GB drives in RAID 5 : if you loose a drive you don't loose anything and you also split the load between the drives. The only disadvantage is that you loose 200 GB of capacity.

I think it's the best choice for HTPC if you don't want to loose any data because you can't really backup the kind of data an HTPC produce (because backuping files over 500MB is a long process - as well as restoring them-, those files cannot be compressed during backup and in the end you have to have a 200GB drive/device to back them up).

Regards,
Stéphane.

Well I'm not talking about types of raid and such. I'm talking strictly about the guys who say a raid 0 array is BAD because if one drive fails then you loose everything. I say IF that happens to a 2-drive 400 GB total raid 0 (for example) YES you loose everything. BUT if you have only ONE 400 GB total drive and you loose IT .. you also loose everything. I'm only talking about the end result. There is no difference. Now the "debate" (which I have debated in other hardware forums) is whether putting TWO drives together will cause either of them to fail more quickly than one drive. "I" say there is a flaw in that thinking. Mainly just from my personal experience with business and non-business clients running raid arrays for several years. Just because you have 2, 4 or 8 drives it does NOT decrease the reliabilty of any of them. The "chances" of a single drive, or a pair of drives or four drives is the same for all drives.

Using that "flawed" that you stand more of a chance of a failure by having say 4 of a particular manufacturer's drives, means that you should never buy a drive from a manufacturer who might sell twice as many drives as his competitor. If you do, you will have more of a chance of failure. Think about it.

[merrypig]

Actually having more drives does decrease the reliability, because in almost all occurences they share a housing and contribute heat to each other....

so n'yer.

[StephaneM]

Originally Posted by Rich A

Well I'm not talking about types of raid and such. I'm talking strictly about the guys who say a raid 0 array is BAD because if one drive fails then you loose everything.

I understand, but you asked "what the difference" between 1 400GB or two 200GB, and I responded that the only difference is performance. That was impliying that I agreed with you for the reliability of 1 versus 2 : it is the same.

Now from a user point of view, I can understand why it is frustrating to loose the whole logical volume when only one drive fail. And that's why if I had two 200GB drives I will surely use them separately and not in a spanned volume, that way you still have 400GB but with greater chances to loose only 200GB (and that's why I talked about RAID 5 right after).

Regards,
Stéphane.

[StephaneM]

Originally Posted by merrypig

Actually having more drives does decrease the reliability, because in almost all occurences they share a housing and contribute heat to each other...

Yeah, but wether they are in use in a spanned volume or not.

[Rich A]

Originally Posted by Eric3a

The stats seem wrong to me.
Because the "M" in MTBF is for "Mean".
No disk fails at it's MTBF. But some fail before, some after.
So by putting 2 drives working together the combined drive will fail at the SHORTEST life of the 2 drives.

Push the reasonning further: If you had an infinite number of drives, and the failure of anyone of them took the whole thing down, you know that the unit wouldn't work because you know statistically at least one drive would be dead on arrival and just wouldn't work.

Some rare disks fail way before their MTBF, others will last a lot longer.
But as you add disks, the actual failure time of the unit becomes that of the shortest life drive.
Which if you're unlucky like my latest raid disk was within 1 hour of receipt.

Perso, for videos I don't care...


But the stats just didn't seem right. Did i get it right?
Eric

heh heh .. I "knew" this would stimulate debate. No I don't "think" you are correct. I probably should have not even used the term mtbf. My whole concept is simply this ...

A guy having a two drive array totalling 400 gb "thinks" he is more likely to have a failure just because he has more drives. First I think that logic is seriously flawed. But that's not really what my concept is about. Rather that it is simply that a drive can fail .. if a drive in a 2-drive 400 gb Raid 0 fails he looses everything. If that same guy had a single 400 gb drive and IT failed, he'd still loose everything. Can we agree on that? Please I'm not talking about Raid 1 or others. Just a striped set Raid 0. When a drive fails .. you loose everything. Don't matter if it's a drive in a stripe OR a single drive. So why obsess about it?

Now the real debatable thing here is the "chance" of any single drive failing.

There are a group of people who say if you have four drives in a raid 0, then the chance of a failure is greater. Sorry I don't buy into that. IF that were true, then since I have 6 computers with about a dozen drives among them I should have a MUCH higher incidence of failure. Right? No! Forget about the "raid" thing for a moment. The "more drives-increased failure" theory should apply to individual drives as well. Remember we all agree that if a multiple drive Raid 0 has one drive fail, you loose everything AS you also do with a single drive. So that is not a bone of contention.

Now if you apply the popular logic that more drives increase failure .. well MAYBE that's true in the grand scheme of things. BUT not for a guy with two 200 gb drives versus a guy with one 400 gb. Given if one manufacturer had an equal quality drive to a competitor and he made 10 million drives, and the other manufacturer only made 2 million drives, your logic would dictate that you should only buy from the guy who made 2 million, because the guy with 10 million is going to have more failures out there.

Look at it this way. Take your four 100 gb drives OUT of the RAID 0 and put each one into it's own computer. Did you just make them MORE reliable? I think not.

[Rich A]

Originally Posted by merrypig

Actually having more drives does decrease the reliability, because in almost all occurences they share a housing and contribute heat to each other....

so n'yer.

Now this is fun. Yeah but having mutliple drives in a raid 0 distrubutes the writing among them, thus reducing the load per drive, versus a single drive having to foot the whole writing load by itself. That alone should increase reliabilty no?

BUT, I agree on the "heat" thing. In fact my latest HTPC is going to be a two unit device. I've moved the raid and default drive/folder drive to an external box. This will GREATLY reduce the internal heat build up of the server.

[Eric3a]

Take your four 100 gb drives OUT of the RAID 0 and put each one into it's own computer. Did you just make them MORE reliable? I think not.

Yes I think you did improve reliability with a system composed of 4 separate computers over the system composed with a computer with 4 drives raid 0.

Basically because one of these 4 computers in the system will fail before the others, but the other 3 will keep running.
At this point you still have 3 computers still running where the 4 raid 0 computer wouldn't.

Comparing the reliability of the system of 4 computers to the 1 computer: The latter fails first.
The first is more reliable.

I agree that the most important part in drive reliability is probably heat. I cook my drives in my current case, and suffer fairly common failures, well under MTBF. My next case will provide plenty of drive cooling and space. Also with larger drive capacities we can make do with less drives. Should improve things a lot.


Eric