HD154UI 3-Drive RAID0 Benchmarks

[Loa]

Hello Mac Husky,

I've only had 2 HD die on me over 20 years+ of using Macs, and both times I've been able to recover all the data myself without using expensive solutions. Both drives were used "as-is", i.e. not in a RAID. (On a side note, drive failures are more common now.)

From what I've read (and others will be able to give you a more detailed explanation), a messed up RAID0 is irrecoverable. Only thing to do, unless you're willing to pay experts to recover your data, is to wipe the RAID and restore from back-up.

As for your question "how to prepare", well, nothing beats having a complete and identical copy of all your data somewhere. Ideally 2 copies: the second one outside your home in case of dire events (fire, power surge, theft, water damage...).

Regarding backing-up your whole RAID0: unless the only data you'll put there is scratch or temporary data, why would you take the risk of not backing-it up???

I've had pretty bad luck with Time Machine, and the fancy "time effect" is all good to help sell the concept, but in actuality it's just eye-candy. My next back-up will be made with the help of the free, exhaustive and proven Carbon Copy Cloner app. It has a LOT more options to customize your back-ups than Time Machine has.

Good luck,

Loa

 

[Tesselator]

I'm confused here: different drives for different data is the best way to improve performance short of a RAID.

So If I'm planning a 3-4 drive RAID, I should be OK with all the examples you gave.

No. But that would be the only way to ensure 100% that there would no access clashes. That was my point. I was also assuming the ridiculousness of that would be overwhelmingly apparent. Right, I mean a different drive for every kind of data? I'm trying to say that in all practicality we just can't avoid every situation where more than one operation is going to compete for drive access at the same time. Mostly it's no big deal and we won't even be able to tell the difference. But when the clashing operations are intense like video editing and such then we might. So don't do that purposefully is all. And if you need to anyway (regularly) for some reason then get a drive just for that.

Then you say that partitions would decrease performance, by making the heads seek data further across the platters.

I haven't tested it but I believe it would, yes. It seems like the logical assumption considering how rotational HDD media works. The farther away (inward or outward) that the two files being accessed are from each other the longer it will take to for the heads to seek from one position to the other. If it needs to do that a thousands time to complete both operations that could take some time and cause undue stress on the drive. Partitioning would ensure that said distance would always be at least from the inner edge of stored data on P1 to the outer edge of P2 - as a minimum - and very often more. If it's all one partition then it may be the case that the two data requests are just one or two blocks apart or something and the seek would be almost instant.

In the end, we arrive at the same conclusion: OS + all your data on the same huge RAID0, partition-less volume.

If you're afraid of messiness or navigation difficulties then make a set of Aliases for the folders and store them on the desktop. Give them an icon that looks like a Disk Volume. ㄖ

...Maybe it's my old habits that are hard to dispel, but having a single volume for everything seems dangerous to me. If something bad happens to that (logical) volume, you're toast. Especially with a hard-to-recover RAID0.

Yup! That's what a back-up is for. Don't run a RAID at any level (0, 1, 5, etc.) without a back-up. RAID redundancy levels and schemes are no substitute for a backup.

By having a RAID0 with a few partitions (4-5) to isolate any volume damage that could occur, how much of a performance loss are we talking about?

Volume damage? I think any such damage that happens to a partition on a RAID will always affect the entire RAID partitioned or not. How much performance loss depends on how/when you access the data on the various partitions. And like I say I haven't tested this so I don't have any real numbers to offer you - just guesses. You will for sure be putting your actuators (the heads moving and tracking across the platter is called actuation) through more hell than with just one partition.

This might help you visualize some of what's being discussed: http://www.metacafe.com/watch/1930741/anatomy_of_a_hard_drive/

✸

 

[Tesselator]

(On a side note, drive failures are more common now.)

I think it's much less common! The percentage of drives failing these days is lower than 5 years ago - would be my guess. Drive engineering, manufacturing technology and etc, has matured a lot.

✌ EDIT: I'm wrong! I looked it up and since the 90's the sophistications in R/W drive head and the increased platter density has reduced overall reliability. Early 90's heads were MR single magnet heads. Since then we've gotten GMR twin magnet heads and today many are 7 to 9 layers of magnets on the heads.

From what I've read (and others will be able to give you a more detailed explanation), a messed up RAID0 is irrecoverable. Only thing to do, unless you're willing to pay experts to recover your data, is to wipe the RAID and restore from back-up.

If you could recover data from the single drive you could probably also recover it from the RAID. If the RAID drive(s) fail completely you also have
to replace the drive before you restore from your back-up. True also for a single drive.

 

[nanofrog]

(On a side note, drive failures are more common now.)

Unfortunately, there's truth to this. The technology has gotten more advanced, but the QC in manufacturing has gone backwards it seems.

The result is causing a greater number of failures.

Research (rather intensive), can help sort out the dogs though.

From what I've read (and others will be able to give you a more detailed explanation), a messed up RAID0 is irrecoverable. Only thing to do, unless you're willing to pay experts to recover your data, is to wipe the RAID and restore from back-up.

Not exactly, but is best to think of it in such dire terms. There's software that's available that helps to recover data. But there's no guarantee that you'll get all of it, or it's uncorrupted. (Check the fine print). This is even the case from data recovery services.

But there are occasions that the mechanical portion of a dead drive are still in tact. By having an identical drive on hand, you can swap out the controller boards between them, and get the drive operational, and the data is still there. This is even one of the techniques data recovery uses. Much cheaper, and faster, if you do it yourself. Assuming of course, the controller board is what died.

In the case of a mechanical failure, you're array is toast. Swap the bad drive for the spare, and rebuild from backups. Redundancy helps significantly, but means you choose a different array type, and accept the compromises.

It really is a good idea to have spares on hand. Though it may seem like a waste of money, it's far cheaper than data recovery. If down time can't be accepted, it's more of a necessity anyway, just as are backups. Neither are areas to be cheap with.

Regarding backing-up your whole RAID0: unless the only data you'll put there is scratch or temporary data, why would you take the risk of not backing-it up???

You should be able to set the specifics in the backup software. Particularly if it's 3rd party.

I haven't tested it but I believe it would, yes. It seems like the logical assumption considering how rotational HDD media works. The farther away (inward or outward) that the two files being accessed are from each other the longer it will take to for the heads to seek from one position to the other. If it needs to do that a thousands time to complete both operations that could take some time and cause undue stress on the drive. Partitioning would ensure that said distance would always be at least from the inner edge of stored data on P1 to the outer edge of P2 - as a minimum - and very often more. If it's all one partition then it may be the case that the two data requests are just one or two blocks apart or something and the seek would be almost instant.

Yup.

Don't run a RAID at any level (0, 1, 5, etc.) without a back-up. RAID redundancy levels and schemes are no substitute for a backup.

Absolutely.

 

[DoFoT9]

Absolutely.

in regards to RAID5, it somewhat already is fairly safe (presuming only one drive fails at a time).. in a home user situation i cant really see more then one drive failing at a time..

if you were REALLY picky i guess you could setup RAID6

 

[nanofrog]

in regards to RAID5, it somewhat already is fairly safe (presuming only one drive fails at a time).. in a home user situation i cant really see more then one drive failing at a time..

if you were REALLY picky i guess you could setup RAID6

In the case of a RAID 5, multiple failures do occur. For example, one disk dies. The array goes into degraded mode, and you get the error. Pop in a new drive, and the rebuild begins. But another disk goes (stress) while it's running. Now your SOL.

It's getting more common it seems as well, as drives, especially consumer models (1E14), have gotten too high a platter density, without any improvements to the UBE ratings.

This is one of the reasons RAID 6 is becoming more popular. Enterprise drives do help (1E15), but neither are entirely immune.

I do get your point with home use and RAID5, but not so much when using consumer models. I've seen too many having dropout issues causing lost data. Very similar to a total failure in the end result. Only the causality was different (TLER values). I don't assume they can be changed, as I've only gotten my hands on one for WD's drives. Who won't even tell anyone that it exists, as they want to sell the RAID Editions instead.

 

[DoFoT9]

In the case of a RAID 5, multiple failures do occur. For example, one disk dies. The array goes into degraded mode, and you get the error. Pop in a new drive, and the rebuild begins. But another disk goes (stress) while it's running. Now your SOL.

so its quite possible for a drive to fail while being rebuilt? that would make sense i guess it would be quite intense with all the calculations going on.

It's getting more common it seems as well, as drives, especially consumer models (1E14), have gotten too high a platter density, without any improvements to the UBE ratings.

UBE as in the encryption they use on them? that can play a part in the life of the drive (encryption would occur before the data is written and after its read, not really causing any extra reads/writes) right?

This is one of the reasons RAID 6 is becoming more popular. Enterprise drives do help (1E15), but neither are entirely immune.

forgive my ignorance but i was under the impression that the main difference between 1E14 & 1E15 was firmware? or was that for a specific case of HD (possibly SeaGate)?

I do get your point with home use and RAID5, but not so much when using consumer models. I've seen too many having dropout issues causing lost data. Very similar to a total failure in the end result. Only the causality was different (TLER values). I don't assume they can be changed, as I've only gotten my hands on one for WD's drives. Who won't even tell anyone that it exists, as they want to sell the RAID Editions instead.

a possible setup for home users could be RAID5, plus a mirror of that RAID5 (RAID51?) - the mirror might have to be a RAID0 though haha! seems to confusing. does that seem feasible or does that have to much redundancy and not enough efficiency/insurance of safe data?

p.s. TLER is WD only? CRC was implemented by SeaGate (from wiki).

 

[Tesselator]

In the case of a RAID 5, multiple failures do occur. For example, one disk dies. The array goes into degraded mode, and you get the error. Pop in a new drive, and the rebuild begins. But another disk goes (stress) while it's running. Now your SOL.

It's getting more common it seems as well, as drives, especially consumer models (1E14), have gotten too high a platter density, without any improvements to the UBE ratings.

This is one of the reasons RAID 6 is becoming more popular. Enterprise drives do help (1E15), but neither are entirely immune.

I do get your point with home use and RAID5, but not so much when using consumer models. I've seen too many having dropout issues causing lost data. Very similar to a total failure in the end result. Only the causality was different (TLER values). I don't assume they can be changed, as I've only gotten my hands on one for WD's drives. Who won't even tell anyone that it exists, as they want to sell the RAID Editions instead.

Yep, and also none of this 6 or 5 covers the pilot error and rogue-ware phenomenon. You know... Where you're trying out this new D2D disk recording audio software and after messing with it for an hour you realize now over half the data on your drives has been trashed. Or you decide that slimming is a good thing to do and now nothing works right. Or you decide to use the new and improved Ultra-Magic-Virtual-Partition-Maker and you end up damaging the partition tables.

For these problems no RAID level is going to save you.

 

[DoFoT9]

intense

 

[Tesselator]

BTW, "U.B.E." stands for Unrecoverable Bit Error rate. It's a rating like MTBR but of course more precise than MTBR and not directly related.

Also for what NanoFroggy was talking about with the RAID6 see "8.2. Atomicity" here: http://wapedia.mobi/en/RAID?t=12.






♨

 

[DoFoT9]

BTW, "U.B.E." stands for Unrecoverable Bit Error rate. It's a rating like MTBR but of course more precise than MTBR and not directly related.

yup i gave it a google and discovered the meaning

how do you know so much got any good references!?

 

[Mac Husky]

No doubt: a suitable backup is indispensable going with a RAID0.
CCC seems to be a good idea to proceed it.

But there are still a few questions to me:

Having a 3 or 4 drive RAID0 and a disk fails:

a) how to find out what disk it is ?
b) how to recover the full RAID0 after having replaced the faulty disk ?

What would be the best preparation for such a "crash"?
I planned to have a clone from my OS X and apps when setting up the Mac.
This clone will be complemented by a regular time machine backup.
Also planning a backup of these backups on an extern disk or NAS.
Will this be enough? Or should I clone the whole RAID0 in fixed time distances?

I don´t want to go for a RAID0 without knowing how to handle it in a worse case of disk failure!!!

 

[Loa]

Hello,

As far as I know, it's the same thing as replacing a regular hard drive: you erase the RAID, reformat with the replaced drive and restore from back-up. Then you're good as new.

Loa

 

[nanofrog]

so its quite possible for a drive to fail while being rebuilt? that would make sense i guess it would be quite intense with all the calculations going on.

Unfortunately, Yes.

UBE as in the encryption they use on them? that can play a part in the life of the drive (encryption would occur before the data is written and after its read, not really causing any extra reads/writes) right?

No, but you did figure out the acronym. I should have listed it out, sorry about that.

a possible setup for home users could be RAID5, plus a mirror of that RAID5 (RAID51?) - the mirror might have to be a RAID0 though haha! seems to confusing. does that seem feasible or does that have to much redundancy and not enough efficiency/insurance of safe data?

A RAID 51 would be two identical RAID 5's, that are then set up as a mirror. Unfortunately, this isn't easy to do, as most controllers don't support it. It's also really expensive if you do manage it, so out of most people's budget I think.

p.s. TLER is WD only? CRC was implemented by SeaGate (from wiki).

Here's Wiki's page on TLER. It's called different things by different HDD manufacturers, but is the same. It has to do with the timings used for recovery. For a single disk, that the OS handles error recovery, the values are 0,7 (read, write; in seconds). For a hardware RAID controller, they get set to 0,0. The values themselves, are stored in the drive's firmware, but there's more than just the TLER values that differentiate consumer and enterprise models. The latter usually has additional sensors at a minimum, and hopefully, better spindle motors and servos. Lately however, I think the motors and servos are identical. The firmware has been adjusted of course.

WD's utility is the only one I've ever managed to obtain. The others are extremely protective of such a utility, as it can allow consumer drives to be modified to function on a hardware RAID controller. Without adjusting these values, they have a tendency to drop out constantly, producing an unstable array.

Yep, and also none of this 6 or 5 covers the pilot error and rogue-ware phenomenon. You know... Where you're trying out this new D2D disk recording audio software and after messing with it for an hour you realize now over half the data on your drives has been trashed. Or you decide that slimming is a good thing to do and now nothing works right. Or you decide to use the new and improved Ultra-Magic-Virtual-Partition-Maker and you end up damaging the partition tables.

For these problems no RAID level is going to save you.

LOL. Very good point.

BTW, "U.B.E." stands for Unrecoverable Bit Error rate. It's a rating like MTBR but of course more precise than MTBR and not directly related.

Also for what NanoFroggy was talking about with the RAID6 see "8.2. Atomicity" here: http://wapedia.mobi/en/RAID?t=12.

I forgot to define it. Thanks for catching it.

Nice link BTW. Hopefully, that will help as well.

 

[nanofrog]

No doubt: a suitable backup is indispensable going with a RAID0.
CCC seems to be a good idea to proceed it.

But there are still a few questions to me:


I don´t want to go for a RAID0 without knowing how to handle it in a worse case of disk failure!!!

a. The system will tell you which disk. Worst case, you'd have to open up the error logs. The location will depend on if you used a hardware or software implementation, but it's rather easy.

b. Locate the dead drive. Pull it, and swap it out for a good one*. You then have to get into where ever you created the array (again is different for hardware vs. software), and recreate a new array. Once created, you pull the data from the clone, followed by any incrementals you may need.

* In some cases, the controller board on the drive is what dies, but the mechanical components are undamaged. When this true, you can swap the controller board from an identical drive (spare). In either case, it's a really good idea to keep spares. Buying an extra identical drive, at, or nearly at, the same time, can allow you to salvage the data on occasion. Saves your data, and some work as well. If you have a spare, but it's not identical, or even of the same family, you won't be able to attempt this. You'd have to replace the dead disk, and recover the data from backups.

Before trusting your data to an array, it's best to experiment with it, so you know exactly what you'd be dealing with if the real thing ever happens.

 

[Tesselator]

* In some cases, the controller board on the drive is what dies, but the mechanical components are undamaged. When this true, you can swap the controller board from an identical drive (spare). In either case, it's a really good idea to keep spares. Buying an extra identical drive, at, or nearly at, the same time, can allow you to salvage the data on occasion. Saves your data, and some work as well. If you have a spare, but it's not identical, or even of the same family, you won't be able to attempt this. You'd have to replace the dead disk, and recover the data from backups.

Here's a playlist I put together that covers this procedure. There are other videos on the topic too if you search.

http://www.youtube.com/view_play_list?p=220FCB17D474C136&playnext=1&playnext_from=PL

 

[DoFoT9]

Unfortunately, Yes.

i didnt mean quite possible, i meant more possible then when the drive is doing nothing.. you understand but haha.

No, but you did figure out the acronym. I should have listed it out, sorry about that.

yup , i didnt read over it too much - just enough to get a slight understanding (clearly not enough at the time)

A RAID 51 would be two identical RAID 5's, that are then set up as a mirror. Unfortunately, this isn't easy to do, as most controllers don't support it. It's also really expensive if you do manage it, so out of most people's budget I think.

ok well that rules that idea out (crosses off To-Do list). is there an "official" way to have a RAID5 and then create a mirror of it? i guess there probably isnt even an unofficial way to do it

Here's Wiki's page on TLER. It's called different things by different HDD manufacturers, but is the same. It has to do with the timings used for recovery. For a single disk, that the OS handles error recovery, the values are 0,7 (read, write; in seconds). For a hardware RAID controller, they get set to 0,0. The values themselves, are stored in the drive's firmware, but there's more than just the TLER values that differentiate consumer and enterprise models. The latter usually has additional sensors at a minimum, and hopefully, better spindle motors and servos. Lately however, I think the motors and servos are identical. The firmware has been adjusted of course.

interesting!! soo.. if the motors and servos (meaning the system that places the head?) aren't updated and you have access to the firmware (some WD) then really, what is the use of paying the extra amount of money if you can do it yourself??

WD's utility is the only one I've ever managed to obtain. The others are extremely protective of such a utility, as it can allow consumer drives to be modified to function on a hardware RAID controller. Without adjusting these values, they have a tendency to drop out constantly, producing an unstable array.

right, because the hard drives wont wait that fraction longer. from the link Tesselator provided i found this in the footnotes, its an ugly document but highlights some of the faults that the 1E14 drives have compared to the 1E15 drives, interesting read.

Nice link BTW. Hopefully, that will help as well.

indeed it does

Here's a playlist I put together that covers this procedure. There are other videos on the topic too if you search.

http://www.youtube.com/view_play_list?p=220FCB17D474C136&playnext=1&playnext_from=PL

HA! thats an awsome set of videos!!! thanks!

 

[nanofrog]

Here's a playlist I put together that covers this procedure. There are other videos on the topic too if you search.

http://www.youtube.com/view_play_list?p=220FCB17D474C136&playnext=1&playnext_from=PL

I don't do anything with video. No camera (video, I can do still shots) or inclination.

I guess I could document on .pdf files, but have a habit of not doing so for personal projects. I've not had enough time, and didn't bother taking the photos.

Are you in the video? Or just did the post production work?

 

[Tesselator]

Hehehe... no no. I do and teach video and CG for work but I had nothing to do with those. I just searched YouTube to find them and created a playlist of them in my channel.

 

[Tesselator]

Here's an instructional video I made: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_Quicktime.mov

Or if that link doesn't then I guess this one: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_Divx.avi (Not so good soundtrack tho)

The AVI plays the best but few people have this free codec installed: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_AVI.avi

 

[Mac Husky]

Thanks nano & Tess - doesn´t sound that difficult as you have a sufficient backup.
Even not much more difficult than to restore a single drive without beeing in a RAID0.

 

[nanofrog]

ok well that rules that idea out (crosses off To-Do list). is there an "official" way to have a RAID5 and then create a mirror of it? i guess there probably isnt even an unofficial way to do it

Nothing officially supports it AFAIK. Some hardware controllers offer 50/60, but not 51/61.

It would be left to experimentation, very questionable, and expensive.

interesting!! soo.. if the motors and servos (meaning the system that places the head?) aren't updated and you have access to the firmware (some WD) then really, what is the use of paying the extra amount of money if you can do it yourself??

There's a few bits of hardware that can differentiate them, such as feed back circuits. Fly height adjustments are one possibility. Also, some consumer drives are produced with UBE's of 1E15, but most aren't.

But, yes, if you have the utility, you can edit the values in an attempt to get less expensive drives to work. There's just no guarantee it will work (stability).

Here's an instructional video I made: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_Quicktime.mov

Or if that link doesn't then I guess this one: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_Divx.avi (Not so good soundtrack tho)

The AVI plays the best but few people have this free codec installed: ftp://ftp.newtek.com/products/LightWave/Videos/ShaderIntro_AVI.avi

I'll have to check these out.

Thanks nano & Tess - doesn´t sound that difficult as you have a sufficient backup.
Even not much more difficult than to restore a single drive without beeing in a RAID0.

No, not really. But that does assume you have the backups, and it actually works. The backup software needs to be tested just as the array does.

I'm trying out Acronis True Image ATM, and am having a few difficulties. I may have messed up the restore from boot media (DVD missing Master Boot Record) , so I need to continue testing. Otherwise, so far, so good. (Vista 64bit & Win7 RC 7100 64bit BTW). No Mac support yet. No attempts yet with Linux.

 

[Mac Husky]

No, not really. But that does assume you have the backups, and it actually works. The backup software needs to be tested just as the array does.

I'm trying out Acronis True Image ATM, and am having a few difficulties. I may have messed up the restore from boot media (DVD missing Master Boot Record) , so I need to continue testing. Otherwise, so far, so good. (Vista 64bit & Win7 RC 7100 64bit BTW). No Mac support yet. No attempts yet with Linux.

I heard of a couple of problems with Acronis before even on PCs.
Insufficient clones and backups with no way to recover data

To test the backup software would mean to generate a RAID0 and backup
the system. Than disturb the RAID0 system and try to restore?! A lot of work
if it doesn´t work Of course less than a crash later on and no way
out of the lost data due to insufficient backup

 

[nanofrog]

I heard of a couple of problems with Acronis before even on PCs.
Insufficient clones and backups with no way to recover data

Most of what I'm aware of, is on the XP side. I didn't see much on Vista 64, but what I did, made me think it might be a good candidate. Then there's Acronis's reputation on the entererpise side...

More testing required.

To test the backup software would mean to generate a RAID0 and backup
the system. Than disturb the RAID0 system and try to restore?! A lot of work
if it doesn´t work Of course less than a crash later on and no way
out of the lost data due to insufficient backup

Yes, it's a massive amount of work, but it's necessary to make sure the backups will be there and function correctly.

It does get tiresome though.

 

[DoFoT9]

pretty jealous, 1.5TB hard drives over here cost about $205Aus ($168US) for a Green Power WD one.. it goes up for there, $400Aus for a 2TB model.

not happy jan!