Hard drive question for Linux/mdadm (Is ERC/TLER required)?

Toggle sidebar Toggle sidebar
H

hasu

Senior member
Apr 5, 2001
993
10
81
Would this hard drive work for MDADM RAID-6? Or should I look for one with CCTL (TLER, ERC)?

TOSHIBA PH3300U-1I72 3TB 7200 RPM
Newegg Item=N82E16822149396

Edit: I have a couple of these in mdadm/Raid-1 config. I am thinking of building a 6x3TB raid-6 array to consolidate all my drives. I heard that mdadm/raid-6 can be pretty slow in writing because of double parity but that is a different story!

Edit: What would be a better choice for mdadm/raid-6?
Item=N82E16822149396
or
http://www.newegg.com/Product/Product.aspx?Item=N82E16822145911


Edit: when I link to Toshiba product page in Newegg, the link automatically changes to something like http://viglink.pgpartner.com/rd.php?r=1636&m=1069073483& what is up with that? See below.

Following link is actually www.newREMOVEegg.com/Product/Product.aspx?Item=N82E16822149396

http://www.newegg.com/Product/Product.aspx?Item=N82E16822149396
 
Last edited:
dave_the_nerd

dave_the_nerd

Lifer
Feb 25, 2011
16,983
1,616
126
I think those drives support it. I have three DC01ACA300s in my NAS, which is basically the same drive, and I can enable CCTL with the smartctl utility.
 
Red Squirrel

Red Squirrel

No Lifer
May 24, 2003
69,718
13,338
126
www.betteroff.ca
Any drive should work as long as it's not a green since they go to sleep, and then the raid thinks the drive failed. I have some toshibas, WD blacks and reds in my various raids.

I *DO* have one with greens, which is where I learned my mistake, the work around is to just have a script that reads and writes data to the volume every second. Kinda dirty. Would not do that mistake again.
 
H

hasu

Senior member
Apr 5, 2001
993
10
81
dave_the_nerd said:
I think those drives support it. I have three DC01ACA300s in my NAS, which is basically the same drive, and I can enable CCTL with the smartctl utility.
Click to expand...

I tried smartctl command to set ERC on a Toshiba (same model) by using the following command and found that it won't any value other than 70,70 or 0,0.

smartctl -l scterc,70,70 /dev/sdx

Is it possible to set say 2 second for erc retry? What is the best value for raid-6?
 
H

hasu

Senior member
Apr 5, 2001
993
10
81
Red Squirrel said:
Any drive should work as long as it's not a green since they go to sleep, and then the raid thinks the drive failed. I have some toshibas, WD blacks and reds in my various raids.

I *DO* have one with greens, which is where I learned my mistake, the work around is to just have a script that reads and writes data to the volume every second. Kinda dirty. Would not do that mistake again.
Click to expand...

I see! I was planning to spin down the drives when not in use (which is most of the time!) using hdparm. Can't mdadm wake them up for reading? That used to work with my old two drive QNAP raid-1.
 
E

Emulex

Diamond Member
Jan 28, 2001
9,759
1
71
Depends on whether the drives fire up reliably and time-out. might have to tweak the timeout to account for a slow starting drive rather than fail the raid eventually.

TLER isn't going to help with linux software raid unless you need to continuosly write to the drive (surveillance)
 
C

CiPHER

Senior member
Mar 5, 2015
226
1
36
Red Squirrel said:
Any drive should work as long as it's not a green since they go to sleep, and then the raid thinks the drive failed.
Click to expand...
That is a misconception. The headparking feature known to desktop drives does not cause disks to be dropped from the RAID array.

Disks are dropped from a RAID array if they exceed the timeout of 10 to 30 seconds depending on controller/firmware. This is caused by bad sector recovery, not headparking!

The headparking takes less than a second to be in operation again, and cannot possibly cause the problems you talk about.

Bad sectors however, can cause havoc on legacy RAID solutions because the dumb firmware cannot distinguish between a harddrive doing recovery (up to 120 seconds) and a failed harddrive. It has all the possibilities to do that (channel reset + reissue I/O) but most of the controllers are not designed that way. This includes the Areca Hardware RAID and other RAID controllers; though 3ware and LSI often are exempt. It also includes Windows FakeRAID (Intel/AMD/nVidia/Promise/Silicon Image/ASMedia) however.

To cope with this problem, TLER is needed. Which does what is says: Time-Limited Error Recovery; otherwise put it limits the time the drive is allowed to perform error recovery. A drive with TLER will return an I/O error after typically 7 seconds, so that it never crosses the 10+ second timeout the controller adheres to.

But for Linux software RAID and ZFS, you do not need TLER. These solutions are properly designed. It can cause faster responsetimes when bad sectors do occur, however, so it may still have some use. But this can also be done with software settings by the operating system; you do not need TLER for this. TLER is only needed if you have a legacy solution working with the disks.

In fact, TLER can also be dangerous. Consider a RAID5 where you lost one disk and thus have no redundancy left. Now you want to rebuild with a fresh disk. Suppose one or more disks have a bad sector, you would want the drive to have all the time it needs to recovery the data, instead of forfeiting it after a few seconds. TLER basically kills your last line of defense: the recovery feature of the harddrive.

WD Greens are perfectly suitable for modern storage; ZFS, Btrfs, ReFS. And also traditional software RAID under Linux/BSD would work fine, though traditional filesystems are not designed for the high uBER-rate of consumer-grade drives. For traditional RAID and filesystems you would want harddrives with 10^-15 or even 10^-16 uBER instead of the usual 10^-14 that consumer-grade drives get. This means enterprise drives will develop bad sectors up to a factor 100 less often when compared to consumer drives, which means the danger of bad sectors can be pushed in the bandwidth where parity can protect against.
 
E

Emulex

Diamond Member
Jan 28, 2001
9,759
1
71
Bad sector recovery typically is 180 seconds per operation, this can cause service failures. I'd just stick to JBOD with consumer drives, raid adds very little protection on consumer gear!
 
Red Squirrel

Red Squirrel

No Lifer
May 24, 2003
69,718
13,338
126
www.betteroff.ca
I've seen head parking knock drives out of an array. I setup a server for the church to act as a backup for the speech recordings and decided to go cheap and go with green drives. Drives kept dropping out left and right. I then realized it's because they were going to sleep. As long as I write to the array once a second I'm ok. I don't really know what the minimum time is but I read that those drives are very aggressive with sleeping so I just do it once a second. Basically it dd's a couple MB worth from /dev/zero then reads it to /dev/null, once a second. Really crude though. I would not do this for my own setup.

I think there's some kind of firmware thing you can do to stop it though. Just don't like messing with that, there's always a risk of bricking a drive if something goes wrong, and the tool is probably some kind of DOS thing that requires a floppy, don't feel like dealing with that. :p

We have an external drive at work we share and it's a WD green and it can take a good 30 to 60 seconds for it to wakeup. That would definitely be enough for raid to mark it as bad. They don't just park the head, they spin down too.
 
dave_the_nerd

dave_the_nerd

Lifer
Feb 25, 2011
16,983
1,616
126
E

Emulex

Diamond Member
Jan 28, 2001
9,759
1
71
I've had great luck with hitachi enterprise drives, IIRC they had 512 byte sectors without emulation in 2TB size which was a big plus!
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom