New SATA RAID Setup - recommendations please

[BlackBox]

This morning I tried to enter my main hard disk.. I heard a terrible marble-rolling-in-a-little-box noise, the disk didn't respond and after waiting a bit, the disk was gone. (Meaning that the pc didn't recognize it anymore.) After boot up all was fine, but I wanted a better solution for my data storage.

I'd want a RAID 1 (edit) and with this solution I want to be able to expand in the (near) future.

So, the 2 disk raid setup my motherboard offers me will not enough. Therefor I was thinking of buying a sata raid card. I'm not familiar with this sort of hardware and a little advice on this concept is welcome. I did find some cards of which I know they are good, Areca... This does get me in a higher price range (and is only PCI-x). After more searching, I found the much cheaper Highpoint cards (and available for normal PCI). Question is, is Highpoint any good? And how about linux support/drivers, any trouble getting those cards to run on linux?

As you might have noticed, the PCI/PCI-x part is an issue. I don't have a PCI-x port on my server, so choosing for Areca will not only cost me more on the card, but will force me to look into some new motherboard and other things as well... Which will be a little overkill, or not? I do want a reliable solution, so maybe this is the only way.

Any other remarks or recommendations will be highly appreciated.

BlackBox

 

[BillParrish]

you make no sense, HD failed, you want a card to do raid 0, you want a reliable solution.

Raid 0 doubles your chance of data loss due to HD failure.

Question is, is Highpoint any good? And how about linux support/drivers, any trouble getting those cards to run on linux?

eh, they work, no more than anyones else.(suse 10.1)

go raid 0+1 or a 3 disc raid 5 + hot spare or 4 disc raid 5 with a spare disc on the shelf if you want reliable protection from HD failure.

http://www.newegg.com/Product/Produ...SubCategory=410&Description=highpoint&Ntk=all

 

[BlackBox]

I corrected my post, sorry for this idiotic mistake. I want raid 1, the mirroring.

About, RAID 5... I thought you'd have to buy a more expensive card to get RAID 5 support?

I also overlooked the obvious... I read this in an other topic on this forum: simply having to large disks and having a program/or script copy everything over night. Cheap as hell but effective enough for me I think.

 

[Asgorath]

Just a little nightmare that just happened to me ealier today..

6 x WD 500GB SATA RE Drives in RAID 5
2 x WD 80 GB SATA Drives in RAID 1
All held in 2 Supermicro 5 in 3 Drive Racks

My Seasonic S12 Power Supply started smoking. Pulled the plug. Got a new power supply. Plugged it back in. All 8 Hard Drives are dead. One of the Supermicro Racks are dead.

And yes, I put on my sadface. Luckily I live about a half hour from the Irvine shipping dock to drop off my RMA drives to WD. Oh, and one more kicker. I wanted to do a WD advance RMA but WD values these drives at like $460 apiece! I would have to put a $460 x 6 hold on my credit card.

Anyways....good luck with that RAID. This sort of thing is a rarity.

 

[BlackBox]

Damn.... you seem to cope with it very well, I would be banging my head against the wall all day... possibly literally.

 

[smartins]

Damn.... you seem to cope with it very well, I would be banging my head against the wall all day... possibly literally.

Exactly. I think if I ever need to switch PSU's or anything else on my soon-to-be raid box, I'll just unplug the disks and controller to make sure that if anything is to fry is at least easily replaceable (hardware).

 

[drizzt81]

Exactly. I think if I ever need to switch PSU's or anything else on my soon-to-be raid box, I'll just unplug the disks and controller to make sure that if anything is to fry is at least easily replaceable (hardware).

I am under the impression that the PSU change was not planned.

 

[Farva]

you make no sense, HD failed, you want a card to do raid 0, you want a reliable solution.

Raid 0 doubles your chance of data loss due to HD failure.

Actually, RAID0 does not double your chance for loss of data, but it does increase it

I have to nit pick

 

[Asgorath]

Damn.... you seem to cope with it very well, I would be banging my head against the wall all day... possibly literally.

The hardware is replaceable. The almost complete configuration of MythTV is what sucks to me. It took me a long time to get Fedora Core 6 fully patched, updated, MythTV installed, Configured. And now this. All my data is off to get wiped. <tear> <tear>

 

[GeorgeHR]

Asgorath ---

Your power supply failed and you want WD to replace your hard drives?

---

It is common to supply a credit card for cross shipping. After all you could just walk away with the hard drives.

$2500 is a small hold.

 

[Michael Daly]

So, the 2 disk raid setup my motherboard offers me will not enough. Therefor I was thinking of buying a sata raid card. [...]
As you might have noticed, the PCI/PCI-x part is an issue. I don't have a PCI-x port on my server, so choosing for Areca will not only cost me more on the card, but will force me to look into some new motherboard and other things as well...

If the 2 disk MB will give you RAID1 now but only limits you for future expansion, use it and upgrade later.

If you use PCI controllers, they are limited by PCI bus speed - that's why you'll want a PCI -X or PCIe card. If you're forced to upgrade the MB, you won't need a separate controller as you can buy a decent MB with built-in SATA controller. Many of them support RAID5 or better. The only issue is how much CPU is required for the RAID5 - some controllers do the calcs (XOR) while others rely on the CPU for that. AFAIK, only the pricier controller cards have built in XOR processing.

 

[Asgorath]

Asgorath ---

Your power supply failed and you want WD to replace your hard drives?

Yes. Yes I do. And replace them they will.

 

[MrGuvernment]

Raid 10 ?

performance of raid 0 and reliability of raid 1 and HD's are so cheap now...

 

[general]

"Actually, RAID0 does not double your chance for loss of data, but it does increase it

I have to nit pick "

It more than doubles it actually. You do get 2x the probability of drive failure. The most common cause of data loss with RAID-0 though is data corruption. That is, the 2x increase in risk due to drive failure gets multiplied by at least 2.00000000000000000001, so you're looking at greatly increasing your chances of data failure.

 

[BlackBox]

If the 2 disk MB will give you RAID1 now but only limits you for future expansion, use it and upgrade later.

If you use PCI controllers, they are limited by PCI bus speed - that's why you'll want a PCI -X or PCIe card. If you're forced to upgrade the MB, you won't need a separate controller as you can buy a decent MB with built-in SATA controller. Many of them support RAID5 or better. The only issue is how much CPU is required for the RAID5 - some controllers do the calcs (XOR) while others rely on the CPU for that. AFAIK, only the pricier controller cards have built in XOR processing.

Yes, I think I'll go down this road. I'm thinking, maybe even forgetting the raid for a while so I don't have that much rebuild-trouble when eventually moving on to a decent PCIx card (with MB). And for now just use a little script to copy modified files from the primary disk to the backup disk over night. Or should I stick with the raid part?

 

[Farva]

It more than doubles it actually. You do get 2x the probability of drive failure. The most common cause of data loss with RAID-0 though is data corruption. That is, the 2x increase in risk due to drive failure gets multiplied by at least 2.00000000000000000001, so you're looking at greatly increasing your chances of data failure.

Just because probability increases doesn't mean it will happen. And no, I don't advocate using RAID0 either

 

[Michael Daly]

Just because probability increases doesn't mean it will happen.

You keep saying that and it remains a meaningless statement. If the probability is less than 1, there is no certainty it will happen. If the probability is greater than 0, there is no certainty it will not happen. Big deal. If the probability increases, the likelihood of it happening increases.

 

[BillR]

You keep saying that and it remains a meaningless statement. If the probability is less than 1, there is no certainty it will happen. If the probability is greater than 0, there is no certainty it will not happen. Big deal. If the probability increases, the likelihood of it happening increases.

I’m not sure about you but I know I’ve been working with computers in various situations since about 1985 and I have yet to meet or see the immortal hard drive.

Your hard disk, all of them will die be it today tomorrow or next year or the year after but the fact is, like any mechanical device it will fail.

As for raid in general, aside from the recent problems with the 680i chipset and bios and all the new problems coming up because it seems no one wants to google any information about Vista and drive configurations, failed raid is just about the number one topic in this forum.

The benefits of raid to the average user, even a heavy duty gamer simply don’t balance out the cost of hardware, marginal performance gain and lost data.

Just my opinion of course, but one I think you will find many old-timers in here share.

 

[BillR]

Asgorath ---

Your power supply failed and you want WD to replace your hard drives?

---

It is common to supply a credit card for cross shipping. After all you could just walk away with the hard drives.

$2500 is a small hold.

Yes. Yes I do. And replace them they will.

I was going to pass this by but simply can't ignore this and put it off.

WD promised you a warrantee for X years against defects in material and workmen ship. The drives you RMA&#8217;d simply don&#8217;t qualify under those terms. It&#8217;s not their fault you chose a less then up to snuff PSU to power their units.

While I don&#8217;t doubt WD will replace the drives because they are highly motivated and customer oriented company this does bring up and ethical issue does it not?

Google &#8220;ethics&#8221; if you don&#8217;t understand that part.

My other issue is simply that bogus warrantee claims such as this keep the prices on product higher for the rest of us.

Not a flame, just pointing out some basic truths

 

[Asgorath]

I was going to pass this by but simply can't ignore this and put it off.

WD promised you a warrantee for X years against defects in material and workmen ship. The drives you RMA’d simply don’t qualify under those terms. It’s not their fault you chose a less then up to snuff PSU to power their units.

While I don’t doubt WD will replace the drives because they are highly motivated and customer oriented company this does bring up and ethical issue does it not?

Google “ethics” if you don’t understand that part.

My other issue is simply that bogus warrantee claims such as this keep the prices on product higher for the rest of us.

Not a flame, just pointing out some basic truths

Yes. That is true. But on that same note, I pay the higher prices that everyone else does too. If a hard drive fails on me for whatever reason, I know WD has my back. That's one reason I choose them to deal with. I build alot of computers and always use either a Seagate or WD drive not because I don't trust Hitachi or Samsung, but because I know a warranty exchange will be fast and easy with either of those companies. In this case maybe I should have made Seasonic buy me new drives. The S12-550W is plenty for the rig I'm building. I called up Seasonic before I purchased and talked to one of their engineers to confirm.

Warranties are there to be used. If a product is not fully backed, I just won't buy it. I have the cell phone store replace my bluetooth headset if it starts acting flaky. I tell the waiter to take my steak back if it's a really crappy cut. In general, I'm one of the few that will actually take advantage of a warranty. But at the same time, I'm a loyal customer. I'll give them business many times over the money I've cost them.

Ethics are good, but sometimes, I just can't afford to be at the top of my ethical game, the world is set up for good people to take advantage of what is put in front of them. If you don't, you just lose. I doubt Seasonic would have picked up the tab on that, even though I just shelled out $138 for their power supply.

 

[drizzt81]

I was going to pass this by but simply can't ignore this and put it off.

WD promised you a warrantee for X years against defects in material and workmen ship. The drives you RMA’d simply don’t qualify under those terms. It’s not their fault you chose a less then up to snuff PSU to power their units.

I understand your point. I think it'd be good if everyone would act "ethically" however let's think about this:

Everyone has to pay higher prices for HDDs since "some" people willingly abuse the lax warranty purpose. Would you say that a "computer illiterate" person should be exempt from requiring the ethical behavior? If Asgorath had not been around to see the smoke coming out of his PSU (and I heard that Seasonic is actually a pretty nice brand) how would he be able to figure out whether it was the PSU that broke the HDDs or a HDD whose malfunction in turn fried the PSU and the other HDDs?

Of course we could spend hundreds of dollars on experts that would try and reconstruct the malfunction and make a recommendation on the most likely cause for the problem. Then we can assign the blame to the company whose product failed. While it is likely the most fair and objective solution to the problem, it is also quite likely more expensive than just replacing the drives under warranty. In the end, the lax replacement restrictions from WD etc. may actually save all of us money.

Back to my initial statement: I do not think that Asgorath should be penalized for actually knowing his stuff. All of us are aware that Joe ``6-Pack'' Schmoe would have just sent the drive(s) back, oblivious to the fact that the PSU was likely the source of the damage. WD would have replaced them. The only difference would be that Asgorath knowledge and experience would be detrimental to his financial well-being. I think that we should not send the message into the world that "If you are clueless you can get away with everything, but if you are knowledgeable, you are screwed".

</soapbox>

 

[Michael Daly]

Actually, RAID0 does not double your chance for loss of data, but it does increase it

I have to nit pick

If you're going to nit pick, you should make sure you're correct.

The probability does double. If you have a RAID0 config, the loss of data will occur if either of the two drives fails. In probability terms that means:
P[failure] = P[Drive A fails or Drive B fails] = P[Drive A fails] + P[Drive B fails]

If P[A fail] = P[B fail], then P[failure] = 2P[one drive fails].

Straight up, the probability of failure of any drive from a manufacturer is the same for a particular type/size/etc. In practice, if two drives are from the same batch, this is most likely true. If they are from different batches, the probability will vary a bit more and the new probability of failure will be twice the average of the two failure probabilities, which, if you're lucky, is less than the prob from the same batch. That's why they recommend using drives from different batches in RAID0 arrays. In practice, the difference is likely to be small.

For RAID1 arrays, the failure that leads to loss of data is the simultaneous failure of all drives. Or:

P[failure] = P[Drive A fails and Drive B fails] = P[Drive A fails]*P[Drive B fails]

Again, if P[A fail] = P[A fail], then P[failure] = P[one drive fails]^2.

So if the failure probability for one drive is one in a thousand (they are much lower than this in a short period of time) then

P[fail] for RAID0 = 2*0.001 = 0.002
P[fail] for RAID1 = 0,001^2 = 0.000001

Clearly RAID1 is a lot less risky than RAID0.

PS - you can confirm the math in any decent probability text book.

 

[Farva]

If you're going to nit pick, you should make sure you're correct.

The probability does double. If you have a RAID0 config, the loss of data will occur if either of the two drives fails. In probability terms that means:
P[failure] = P[Drive A fails or Drive B fails] = P[Drive A fails] + P[Drive B fails]

If P[A fail] = P[B fail], then P[failure] = 2P[one drive fails].

Straight up, the probability of failure of any drive from a manufacturer is the same for a particular type/size/etc. In practice, if two drives are from the same batch, this is most likely true. If they are from different batches, the probability will vary a bit more and the new probability of failure will be twice the average of the two failure probabilities, which, if you're lucky, is less than the prob from the same batch. That's why they recommend using drives from different batches in RAID0 arrays. In practice, the difference is likely to be small.

For RAID1 arrays, the failure that leads to loss of data is the simultaneous failure of all drives. Or:

P[failure] = P[Drive A fails and Drive B fails] = P[Drive A fails]*P[Drive B fails]

Again, if P[A fail] = P[A fail], then P[failure] = P[one drive fails]^2.

So if the failure probability for one drive is one in a thousand (they are much lower than this in a short period of time) then

P[fail] for RAID0 = 2*0.001 = 0.002
P[fail] for RAID1 = 0,001^2 = 0.000001

Clearly RAID1 is a lot less risky than RAID0.

PS - you can confirm the math in any decent probability text book.

I will not argue the math of probability. However, if drive A is ok and drive B is ok, along with the RAID controller not failing, then the RAID array stays intact. So as you can see, theoretical failing vs something actually failing are not one in the same.

 

[daddybigwords]

If you're going to nit pick, you should make sure you're correct.

The probability does double. If you have a RAID0 config, the loss of data will occur if either of the two drives fails. In probability terms that means:
P[failure] = P[Drive A fails or Drive B fails] = P[Drive A fails] + P[Drive B fails]

If P[A fail] = P[B fail], then P[failure] = 2P[one drive fails].

Straight up, the probability of failure of any drive from a manufacturer is the same for a particular type/size/etc. In practice, if two drives are from the same batch, this is most likely true. If they are from different batches, the probability will vary a bit more and the new probability of failure will be twice the average of the two failure probabilities, which, if you're lucky, is less than the prob from the same batch. That's why they recommend using drives from different batches in RAID0 arrays. In practice, the difference is likely to be small.

For RAID1 arrays, the failure that leads to loss of data is the simultaneous failure of all drives. Or:

P[failure] = P[Drive A fails and Drive B fails] = P[Drive A fails]*P[Drive B fails]

Again, if P[A fail] = P[A fail], then P[failure] = P[one drive fails]^2.

So if the failure probability for one drive is one in a thousand (they are much lower than this in a short period of time) then

P[fail] for RAID0 = 2*0.001 = 0.002
P[fail] for RAID1 = 0,001^2 = 0.000001

Clearly RAID1 is a lot less risky than RAID0.

PS - you can confirm the math in any decent probability text book.

You're wrong. Suppose the probability of drive A failing in the first year was 50&#37;. Does that mean that if you have two drives, the probability is 100%? No, of course not. There's a 50% chance that drive A won't fail. There is also a 50% chance that drive B won't fail. That gives you a 25% chance that neither drive will fail.

P[RAID0 survive] = (1-P[A fail]) * (1-P[B fail])

"PS - you can confirm the math in any decent probability text book."

 

[Michael Daly]

theoretical failing vs something actually failing are not one in the same.

Again, you only demonstrate that you don't know anything about probability.

 

[Asgorath]

Again, you only demonstrate that you don't know anything about probability.

Theoretical failing vs. Actual Failures should be near one and the same if the theoretical failures take all variables into consideration. Here, I think many variables are being overlooked, and thus actual failure rates are much different.

A model is only as good as the formulas and input that is put into it.

 

[Farva]

You're wrong. Suppose the probability of drive A failing in the first year was 50%. Does that mean that if you have two drives, the probability is 100%? No, of course not. There's a 50% chance that drive A won't fail. There is also a 50% chance that drive B won't fail. That gives you a 25% chance that neither drive will fail.

P[RAID0 survive] = (1-P[A fail]) * (1-P[B fail])

"PS - you can confirm the math in any decent probability text book."

Theoretical failing vs. Actual Failures should be near one and the same if the theoretical failures take all variables into consideration. Here, I think many variables are being overlooked, and thus actual failure rates are much different.

A model is only as good as the formulas and input that is put into it.

Again, you only demonstrate that you don't know anything about probability.

I seem to have 2 other people agree with me. So I guess we 3 are wrong and you are right.

 

[mutazx]

So.... SATA RAID recommendations?

 

[Farva]

So.... SATA RAID recommendations?

What do you want to do with it? Is data extremely important?

 

[Madwand]

Yes, I think I'll go down this road. I'm thinking, maybe even forgetting the raid for a while so I don't have that much rebuild-trouble when eventually moving on to a decent PCIx card (with MB). And for now just use a little script to copy modified files from the primary disk to the backup disk over night. Or should I stick with the raid part?

It should not be a question of RAID vs. external backup. It should be external backup first, and then perhaps RAID. The two are largely complementary, and an external backup is logically superior to just RAID as long as it's current.

PCI-X is generally backwards-compatible with PCI (but always check the details), giving a potential alternative. However, IMO a better approach might be to upgrade the motherboard first, and trying its on-board RAID, while leaving the option for an add-on controller down the road. For home users, it's generally going to be better to go PCIe instead of PCI-X because the former is easily available in the consumer space, and the latter is targeted towards higher-end servers, and moreover even some of them are going PCIe.

E.g. 140 MB/s 3-drive RAID 5 write performance using on-board nVIDIA, nForce 430:

Absolutely no standard PCI RAID controller will give this performance, regardless of cost, just due to the limitations of the (standard 33,32) PCI bus. If you add PCI gigabit to that setup, you're going to limit the performance even further.

D:\tools>xxcopy /y f:\test\test0\10.gb n:\test\test9

XXCOPY == Freeware == Ver 2.93.1 (c)1995-2006 Pixelab, Inc.
...
-------------------------------------------------------------------------------
F:\test\test0\10.gb 10,000,000,000
-------------------------------------------------------------------------------
Directories processed = 1
Total data in bytes = 10,000,000,000
Elapsed time in sec. = 71.19
Action speed (MB/min) = 8429
Files copied = 1

Add-on storage controllers typically have a richer feature set and better support, but nothing's perfect and anything + backup is better than relying on just the storage controller, however good, to preserve your data. Moreover, much of what's sought with an add-on storage controller can be obviated by a good backup.

 

[drizzt81]

E.g. 140 MB/s 3-drive RAID 5 write performance using on-board nVIDIA, nForce 430:

Wow, that is extremely impressive. What drives are those? What OS are you running? My 3 drive, WD5000YS array is significantly slower than that. Either nVidia's solution is a lot better than Intel's ICH8r's implementation or the drivers for Win2K3 x64 stink.

I will not argue the math of probability. However, if drive A is ok and drive B is ok, along with the RAID controller not failing, then the RAID array stays intact. So as you can see, theoretical failing vs something actually failing are not one in the same.

I do not think anyone has argued what the actual values probabilities of drive failure is. In fact, the actual value is pretty much irrelevant for the probabilistic calculation. Also, since we are talking about "reliability" issues again, I think it is safe to say that the majority of the [H] forum readers do not have a sufficiently large sample of observed HDDs and RAID arrays to draw any type of conclusion about the reliability.

You're wrong. Suppose the probability of drive A failing in the first year was 50%. Does that mean that if you have two drives, the probability is 100%? No, of course not. There's a 50% chance that drive A won't fail. There is also a 50% chance that drive B won't fail. That gives you a 25% chance that neither drive will fail.

P[RAID0 survive] = (1-P[A fail]) * (1-P[B fail])

"PS - you can confirm the math in any decent probability text book."

As you pointed out, the math may have been off. I believe that this topic was previously visited in these posts:
http://hardforum.com/showpost.php?p=1028839595&postcount=165
http://hardforum.com/showpost.php?p=1028840390&postcount=168

Depending on whether we assume that drive failures are independent, mutually exclusive, both or neither the correct answer will be different.

I seem to have 2 other people agree with me. So I guess we 3 are wrong and you are right.

Actually, daddybigwords did not agree with you. He merely remarked that he considered Michael Daly's probability calculation to be incorrect. If we are to believe masher:

[...]In this particular case, if the chance of one drive failing is n, the chance either or both failing [is] 2n-n^2.

However, for very small probabilities such as drive failure rates, the difference between [2n-n^2] and 2n is very small.

 

[Madwand]

Wow, that is extremely impressive. What drives are those? What OS are you running? My 3 drive, WD5000YS array is significantly slower than that. Either nVidia's solution is a lot better than Intel's ICH8r's implementation or the drivers for Win2K3 x64 stink.

2 x Maxtor 7V300F0 + 1 Seagate 7200.10 320 GB. This particular combination has been carefully selected for optimal performance (not!).

Stripe size = 32k.

Vista RC1. <- This might help with nice big request sizes (and so make the stripe size less important). Unfortunately painful to get another OS running again on the same machine for comparison, but I should do that..

Semi-random XP + Vista nVIDIA driver installations until I got something that worked.

I was very surprised by this performance as well; I'd seen much worse earlier and given up on nVIDIA RAID 5. I heard that some people are using on-board nVIDIA RAID 5 for uncompressed HD recording, and that made me go "huh?" and re-test, and see this result.. I dunno what changed since the last time I tested this. Perhaps a combination of better drivers + Vista + lack of whatever mistakes I might have made earlier.