Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[m2k3423]

Hi,

I read some posting here talking about Western Digital Caviar SATA drives, mostly focusing on 4K sector (so called advanced sector).   However, the bigger problem is with the implications of Western Digital hardwiring TLER to disable state  (TLER = Time-Limited Error Recovery) in recent shipments.

With TLER permanently disabled, whenever there is a sector read/write error, the drive will automatically try to recover from the error (perhaps by re-mapping).   TLER if enabled is supposed to instruct the drive to time-out after a specified time and return control to the drive controller. 

The TLER behavior has deep implications on RAID. If disabled, the RAID controller (software or hardware) will timeout waiting for the drive whenever it encounter problem(s) with sectors while the drive attempt aggressively trying to recover (remap). The consequence will be RAID dropping the drive, and this will cause the RAID array to loose synchronization. For a large capacity RAID array, this may means hours and hours of re-sync.  HOWEVER, this is just a inconvenience.

Recently stress testing two WD15EARS (1.5TB) before inserting them as RAID drives. What I found is that after continuously writing for 12 hours, both drives consistently went into what I called "tail-spins"; i.e. they seems to be non-contactable anymore via the SATA interface.  I suspect it went into some kind of automatic error recovery.  After recycling power, I read the SMART information, and nothing was reported about read re-rty or sector remap.  The drive behave normally on re-cycling power. Now, this gave me reason tosuspect that the Western Digital drives are not reporting SMART information honestly.   One obvious sign is that the writing seems to get slower as the test progresses.

I must mention that Western Digital do have Caviar drives specifically for RAID in the RE series, which cost more.

I am now made to believe that the Caviar Green drives are lower-cost lower-performance drives, by means of having the electronics and firmware do more error recovery to compensate for lower quality disk media.  Perhaps, this is why TLER must be disabled, and SMART information is strangely not reflecting the historic behavior faithfully.

Therefore, it is not wise to use WDC Caviar Green for building RAID array on DNS323, not because of 4K sector causing low performance, but the error re-tries which kills performance, and the suspected implications of inferior disk surface quality. 

P/S: I did the same on a Seagate 1.5TB drive,  it is a hotter drive, but, reliable drive. The drive support ATA SCT ERC command, the equivalent of WDC's TLER.

[scaramanga]

Please see my post here: http://forums.dlink.com/index.php?topic=12977.msg85019#msg85019

[tentimes]

Turns out it is easily fixed and just needs implemented in BIOS. TLER os less important than initially thought.

[m2k3423]

Well, the main point is: the retry behavior of Western Digital Caviar slows down performance significantly, as compared to a Seagate drive. Though a sample size of 2 Caviar 1.5TB drives are not statistically significant, but both have exactly the same bad performance on the same computer (same CPU, same memory, same SATA controller, ... everything same).

To observe the Caviar getting itself into seemingly infinite retries, just plug the Caviar into a USB-SATA docking bay, connect the docking bay to any Linux machine and run badblock with options -w -s -v. For 1.5TB partition, it will take about over 10 hours for single pass of write-verify, and it will take 4 passes for each test pattern. 

For me, both the Caviars 'failed' after the first pass, 'fail' as in it got into infinite re-tries (drive activity light stays on forever). For Seagate 1.5TB SATA drive, no problem at all, except that the drive servo voice-coil is significantly audible as compared to the Caviars.

In addition, when the drives are put into RAID arrays, the Caviars can do at best 65Mbps when synchronizing, while the Seagate can sync at sustained 75Mbps.  This is not in-theory, but real measured numbers.  You can try repeating the tests and share if the same observations are true. 

More importantly, the Seagate drive allow configuring the TLER/ERC using  smartctl -l scterc,70,70 to set read and write timeout to both 7 seconds (note: you need to download the bleeding edge version of smartmontools to have this command option).  I did read the article pointed out, but as far as what I can observed for the kernel version in the DNS323, the TLER/ERC did made some difference.  I did observed the Seagate drive doing a re-try when performing RAID re-sync, dropping to 20Mbps+ for a few seconds, while it is audible that the head was retracting. However, upon completing re-sync, I read the SMART info and did not find any re-mapped sectors. More importantly, the recovery was really quick, within 15 seconds, most importantly, it did not trap itself in infinite re-tries.

One more point to note is that the temperature of the drives are normal - well ventilated, stayed at 41C most of the time.

[gunrunnerjohn]

Moral of the story?  Don't buy WD desktop drives.

[tentimes]

At least until dlink update the firmware I'm not holding my breath though.

[gunrunnerjohn]

Since it appears to be a conscious decision on their part to restrict the use of those drives in enterprise type applications, I wouldn't plan on it being changed anytime soon.