First build, 16TB raw

[keenan]

Request Queuing
I don't buy that one. If the enterprise drive does indeed perform better in some IO task is probably because WD just disabled some feature in the consumer drive. It would be downright bad management / poor engineering to have different electronics for the two lines of drives given the benefits of sharing it between the lines.

I might buy that their queuing is optimized more for high IOPS in a random usage pattern versus high throughput in a streaming pattern on the consumer side. Or they could intentionally use a naive scheduler on the consumer drives to artificially increase performance of the enterprise models. Utilizing NCQ certainly doesn't seem to make much difference on consumer drives, while it seems like it should.

 

[okashira]

I might buy that their queuing is optimized more for high IOPS in a random usage pattern versus high throughput in a streaming pattern on the consumer side. Or they could intentionally use a naive scheduler on the consumer drives to artificially increase performance of the enterprise models. Utilizing NCQ certainly doesn't seem to make much difference on consumer drives, while it seems like it should.

Yeah... it would be reasonable to have the NCQ tuned differently for the different lines.

 

[dividebyzero.pro]

Usually an engineered solution has some measurable standard. You seem to just make up stuff.

For example:

My performance metric is to have 14 days of TV recording space on my hard drives. I think that is about 100 hours. When I get near 100 hours of free space left, I either delete some recordings or buy more space.

12TB of media seems to be around $1000 for drives. About $100-200/year in lost income. Seems that the recommendation to use Netflix is cost competitive.

What have I made up? Most of this thread is asking advise about my intended goal, and discussing options. No, Netflix is not cost competitive, as they won't hold any VM images, or a database, or any of the other data that I will be storing. Only one facet of this project is to hold media for streaming (and that doesn't just mean movies, or whatever else Netflix has)

Either you didn't understand the thread, didn't read it, or you are trolling? As your suggestions (GeorgeHR and Archaea) have nothing to do with building this box. Please avoid replying unless it is to discuss this box or something related.

 

[dividebyzero.pro]

WD Green drives also support TLER but default to disabled; you do not need RAID edition harddrives.

Also, using TLER will cause misbehaving HDDs to be kicked out of the array even faster - and NOT PREVENT IT! TLER means Time-Limited Error Recovery, and it will LIMIT the time it spends on recovering bad sectors. As a result, the RAID engine would get a I/O error on the drive much sooner and kick it out of the array much sooner.

In an enterprise, mission-critical environment, that's what you want. You don't want your important server to freeze for 30 seconds or even longer. If any disks even but scratches its bum, you want to throw it in the waste-bin and replace with any of the hundred reserve disk lying on your desk. Half a minute downtime can mean millions of damage in some cases.

So TLER is useful for keeping the array responsive during individual HDD errors; not for preventing them to drop out of the RAID array! Using TLER in RAID0/JBOD configurations is also strongly not recommended; you only increase the risk of data loss.

TLER is not very well understood by its users. Virtually all home users do not need it.

By the way, have you considered ZFS?

Yes, I am on the path of building this box out for ZFS storage now. Everything looks to be on track, just waiting for parts to arrive now.

 

[sub.mesa]

No you can't. It's been disabled on new WD consumer drives (Blue, Black, Green) for some time now. Even the ATA-8 SCT commands are disabled on the newer firmwares, e.g.:

Quite possible, but then existing GP disks don't have to upgrade firmware and can continue to use TLER, if people wanted. I guess WD figured they need TLER as argument to make people buy more expensive disks, while its just a simple software toggle that they simply hide/make inoperative on the consumer-grade versions.

But if what you say also applies to current WD Green 1.0/1.5/2.0 disks then TLER is indeed only for the more expensive RAID-edition drives.

Though I agree with the rest of your analysis. TLER will prevent the array from freezing for more than a few seconds, and I think some 'enterprise' RAID cards might do a poor job of dropping disks due to timeouts (assuming they'll fail out on their own and return an error).

I agree that the response on long I/O time(out)s depends per controller or RAID-engine. But if using ZFS the OS will handle I/O errors, and generally non-windows OS handle those quite fine. ZFS will also record how many times a read or write request failed; and if it can't access the disk for a read request now, it will read from redundant data if available (mirror/raid-z/copies=2).

Having TLER disabled is generally good, since you will allow the drive to try hard to recover the existing data at a particular weak sector, so it can swap it out with a reserve sector. It won't do that until it can read (recover) the existing data - OR - when the OS writes to that sector location, then the old data is not needed anymore and the disk writes to a reserve sector instead; swapping it out and fixing the issue.

On some forums, i've heard they noticed recovery times of up to 53 seconds for consumer-grade drives. That means that with TLER at < 10 seconds, it would prevent repairing such damage and actually may cause data-loss, for two reasons:

1) with non-redundant arrays (RAID0) TLER may prevent data recovery which was otherwise recoverable - you just didn't allow it long enough to try to get your data back
2) with redundant arrays, TLER may cause a disk dropout (REMOVED/DETACHED/FAILED) and after replacing the disk a rebuild has to occur; during this the array is in RAID0-mode - the redundancy protection is gone. Any other errors on existing disks may cause a failed/forked array. Recovery in such situations is still possible but requires proper action by the user.

With TLER disabled and use of ZFS and properly increased the timeout value to at least 60 seconds, this would allow the disks to repair all recoverable weak sectors. All you would notice is a hick-up of maximum a minute, that's all the patience you need to have to allow the drive to fix itself and continue normal operation without degraded arrays and such. For businesses, this may be unacceptable; for consumers, i think its the smartest choice.

 

[GeorgeHR]

What have I made up? Most of this thread is asking advise about my intended goal, and discussing options. No, Netflix is not cost competitive, as they won't hold any VM images, or a database, or any of the other data that I will be storing. Only one facet of this project is to hold media for streaming (and that doesn't just mean movies, or whatever else Netflix has)

Either you didn't understand the thread, didn't read it, or you are trolling? As your suggestions (GeorgeHR and Archaea) have nothing to do with building this box. Please avoid replying unless it is to discuss this box or something related.

You used a form of the word "engineer." My comment was that you did not do any engineering. I gave an example of one engineered solution as well as the math behind it.

If you want 16TB of space, just go buy 16TB of space. But don't confuse that with engineering.

----

Most of my business computing engineering is in preserving 100GB of data a month. I used to buy enough hard drives to hold at least a year's worth of data at a time. Now the best solution is to buy the cheapest ($/GB) hard drives in multiples of 4.