WD offers new Red HD Line for RAID

[John.Murray]

One of the biggest issues I've had with "prosumer" RAID devices, is the fact they are usually populated with desktop class hard drives.  If one of the member drives encounters a CRC error, it can go into a recovery mode lasting up to 7-10 seconds.  This is enough for the host RAID device to mark that drive as failed.

In the case of Western Digital, it's been possible to update the firmware (unsupported by WD, of course) to shorten CRC recovery time to match their enterprise class drives.

Now they have a new class of drives specifically designed for RAID:

http://www.wdc.com/en/products/products.aspx?id=810

[tived]

thanks for sharing that John
looks interesting
Henrik

[Ellis Vener]

A review of the new Red Drives from WD: http://www.pcper.com/reviews/Storage/Western-Digital-Red-3TB-SATA-SOHO-NAS-Drive-Full-Review

[chrismurphy]

In WDC drives, the feature is TLER. What amusing about this, to me, is that disabling the error correction is the enterprise feature you pay extra for. Instead of correcting the error, the disk reports the error, and a hardware RAID controller of sufficient capability will simply rebuild the data for that failed sector (or stripe) with parity. So the disk isn't considered failed, just a particular LBA and hence stripe. But as you say, if the feature is not-enabled, i.e. ECC is fully enabled, possibly 1/2 minute to full minute delays are possible with some errors, and the disk is effectively non-responsive while it's working on that. Hardware RAID controllers may consider such a drive as failed, and drop it out of the array even though 99.999% of the LBAs are completely readable with no errors.

Note that software RAID does not have this "problem" it's much more tolerant data delays from drives before they're dropped from the array. And many consumer RAID products are in fact software RAID. The problem with such proprietary solutions is that it's not clear what they're using, even affordable disreet RAID cards may be in effect running software RAID on the card and thus don't need TLER capable disks.

[Dustbak]

Yep, another nice detail is that up until several years ago every WD HDD was able to turn off TLER making every WD HDD suitable for RAID usage. As a 'user option' WD disabled this possibility so they can sell you the much more expensive RAID capable drives.

I now have a RAID consisting of the black caviar drives and one of Hitachi 7200rpm drives. I prefer the Hitachis, faster and quieter.

[tived]

Hi guys
What drives are preferred for raid these days?

It's all getting a bit confusing

Thanks
Henrik

[Dustbak]

WD HD are good drives, I find them fairly noisy and a bit pricey though. You also need to get them with TLER disabled (eg. The Black caviars or the RE versions).

I prefer fast spinning drives (7200rpm min.) The Hitachi 7K3 drives I find a good alternative for the WD's. cheaper, faster and more silent.

I have had a bad bunch of black caviars but aside that none of these drives have failed me sofar. I mostly use RAID5 and am not that bothered when 1 drive gives (though RAID6) would be nicer andore failsafe.

[tived]

Thanks,

I am looking at replacing my two IBM M1015 controllers and ad an Areca 1882xi 16-24 channel controller.

I currently have some cheaper Seagate SV35 disk (8x) but these are not very good - just cheap and cheerful - I already have had one have bad sectors, which meant I had to pull the disk in my R0.

I am wanting the Areca, so that I can run RAID-6 over 12 disks, to give it the speed but also the redundancy (2 failed drives)

Henrik

[John.Murray]

Henrik - the Red's will work well in your application.

I think there's some confusion regarding Consumer and Enterprise class drives and RAID.  All drives have error recovery; it is not possible to disable it.   TLER stands for Time Limited Error Recovery, in the case of WD - the RE, Velociraptor and now, Red have TLER - which limits the error recovery time to 7 seconds - making them compatible with embedded (hardware) RAID controllers.

By the way, TLER is WD's term, Samsung and Hitachi use CCTL, Seagate uses ERC - all refer to the drives SCT Error Recovery Control Setting in firmware.

There is a nice utility that will give this information (and *much* more):

http://smartmontools.sourceforge.net/man/smartctl.8.html

smartctl -l scterc /dev/sda

smartctl 5.42 2011-10-20 r3458 [x86_64-linux-2.6.32-279.2.1.el6.x86_64] (local build)
Copyright (C) 2002-11 by Bruce Allen, http://smartmontools.sourceforge.net

SCT Error Recovery Control:
Read: 70 (7.0 seconds)
Write: 70 (7.0 seconds)

It happens to be included on this live boot .iso of freeBSD

http://mfsbsd.vx.sk/

update:  I discovered a Windows Version, unfortunately the Graphic UI does not reveal SCT ERC values.  Fortunately smartctl.exe can be found in c:\program files (x86)\gsmartctl\  download it here:

http://sourceforge.net/projects/gsmartcontrol.berlios/files/gsmartcontrol-0.8.6-2.exe/download

C:\Program Files (x86)\GSmartControl>smartctl -l scterc pd3
smartctl 5.41 2011-06-09 r3365 [i686-w64-mingw32-2008r2(64)] (sf-win32-5.41-1)
Copyright (C) 2002-11 by Bruce Allen, http://smartmontools.sourceforge.net

SCT Error Recovery Control:
           Read:     70 (7.0 seconds)
          Write:     70 (7.0 seconds)

[chrismurphy]

You also need to get them with TLER disabled (eg. The Black caviars or the RE versions).

The enterprise versions may have TLER enabled by default, the consumer Black drives have it disabled.

It's all getting a bit confusing

It is confusing.

Consumer drives will make multiple correction attempts which can take quite a long time (maybe a minute). More sophisticated hardware RAID expect a drive to quickly correct for errors, or quickly tell the RAID controller there's been a sector error. Upon an error being delivered such a RAID controller will rebuilt that sector from parity. Consumer drives taking a long time to error correct is interpreted by the controller as a disk failure and it removed the drive from the array.  For RAID 5 or 50, there is a good chance of losing the entire array while it takes time to rebuild that one drive, as another drive encounters a minor error - the controller drops a second drive and poof the array is totally inconsistent and almost certainly not recoverable at all.

mdraid (linux software RAID), ZFS, btrfs, and probably ReFS are not this fussy. They'll wait for the disk to error correct instead of dropping the drive from the array. So you can use consumer disks in such setups. The problem is that there are many implementations of "hardware RAID" that actually are just embedded linux mdraid. So it's not really clear what kind of disk capability you really need to pay for without thoroughly understanding how your RAID works - or you really trust the advice of your RAID vendor.

[tived]

Thanks guys,

I am probably side tracking here, but I think it can still be beneficial to the community here.

I use RAID for two things, one for increase throughput R0, and I am wanting to set up a large array for my images that I can access quickly and keep adding to, modify etc.

I guess the later doesn't have to be local

Locally, on my PC, I have currently set up 3 raid volumes on three different controllers, all R0, for speed. Here I am running the risk of drive failure. Data is backed up to some extend.
The three volumes are for OS/APPS, TEMP and finally IMAGE files to be worked on.

On the hardware level it's
4x Intel 520 SSD for OS onboard controller
8x Intel 520 SSD for TEMP On IBM M1015
8x Seagate SV-35 for Image to be worked on IBM M1015

Those Seagate are not the greatest drives and will be retired to other duties when I finally get to rebuild all this

I have some 30 TB of images scattered on drives around the office and I am wanting get all this together in one place.

I am thinking of one or two Areca 1882 controllers, I am not sure if these are about to be replaced with SATA4 or PCIe3, they may very well be by the time I get around to do this.

What would be a good strategy?

Thanks
Henrik

[John.Murray]

Henrik:  The Areca is a great controller.  I tend to shy away from going after "one big honking" drive (ie: RAID 5 6 or variants), unless you have a compelling reason to.  Drive pricing has come down to the point where putting together a RAID 10 (more properly 1+0) config makes some sense.

Why do I avoid 5 or 6?  It comes down to parity; the overhead of calculating the parity value, and the risk associated with something bad happening while parity is being written out (write hole).  If you choose RAID 5 or 6 be sure you select the battery backup option on your controller and test it.

Also consider time lost in the event of a rebuild...... - You'll lose your array if a 3rd spindle fails during a rebuild

You can put together a RAID 10 of similar capacity for not that many more spindles (12TB=8 for RAID 10, 6 for RAID 6)) and in the case of a failed spindle - the rebuild is a simple copy with no parity overhead......  The scenario where you could lose your array would be a failure of both spindles comprising one of the RAID 0 pairs.....

A few years back there was an excellent discussion between George Ou and Robin Harris about the merits and pitfalls of both:

George: http://www.zdnet.com/blog/ou/comprehensive-raid-performance-report/484

Note that his tests are using Intel ICH8 onboard RAID - these are *not* indicative of the result you'll get using an Areca or Promise controller

Robin: http://www.zdnet.com/blog/storage/raid-10-is-the-cadillac-of-raid/131

[tived]

Thanks John,

I am not particularly hooked on RAID-5/6, I am just looking for a good compromise where I have good access and good "security/safety". The down time is certainly an issue, as i usually always happens when you have to rush something ;-)

Areca or LSI Some of the guys who also have my particular mainboard have experienced difficulties with the otherwise excellent LSI 926x/928x controllers. So I was thinking that if I got a large Areca 1882xi-24 with 4GB Cache and battery backup for it I would be set for a little while.
One thing that I am condemplating is if two are better then one controller, that maybe getting 2 12 or 16 channel controllers would be a better option and spread the load out over multiple controllers? Do you have any experience with this?

I am also looking at a backup solution like one of the Synoptic 12-disk drives, that can be stacked? or similar solution, again safety and speed compromise is the key here ;-)

Then ones I finally get the controllers which drives are the most suitable for the task

thanks very much for sharing your insight, its much appreciated.

Henrik

PS: I will have a look at the links now - thanks

[John.Murray]

Although expensive, going with 2 controllers would be ideal in a RAID 1+0 configuration; each side of the member RAID 0 pairs would be serviced by a different controller.  

We configure Promise Cabinets using dual SAS controllers in exactly this fashion.

I'd strongly recommend contacting Areca directly and talking your options over with an engineer - Pre-Sales tech support is taken seriously by all major storage vendors, take advantage of it!

[tived]

Thanks for your advise John, its much appreciated.

Its the expense that stalling this problem at the moment, which may then ask the question - is it then needed if you can't get the cash for it ;-)

The main purpose of this machine is Panoramic stitching and retouching workstation.

Henrik