SATA Drive Dying??

[klees]

I have two 120G Samsung SP1213C SATA drives and I believe one of them is acting weird. During boot when the modules are loading I now get a scsi_wait_scan loaded message that I don't recall seeing before. Moreover, when I perform a df -h a FAT32 partition I have on the drive takes longer to show up than the other partitions even though the partition is empty. I reformatted the drive and re-partitioned it again and I still get the aforementioned.

Are there any diagnostics/commands I can run to determine whether the drive is behaving weird? I would like to be ready ahead of time if in fact is dying. What about testing the drive's performance? Thanks in advanced.

[Rob1n]

You can try smartmontools - that uses the inbuilt SMART diagnostics which may tell you whether there's any errors - after emerging do:

Code: Select all

smartctl -d ata -a /dev/sda

to get a full report and

Code: Select all

cmartctl -d ata -t long /dev/sda

to start off a full test (results will be show in the full report, or see the manual pages for how to check progress and view only test results).

For testing performance you could use "hdparm -tT /dev/sda" to get a ballpark figure. A proper testing tool like bonnie++ will give you more realistic results though.

[klees]

You can try smartmontools - that uses the inbuilt SMART diagnostics which may tell you whether there's any errors - after emerging do:

Code: Select all

smartctl -d ata -a /dev/sda

to get a full report and

Code: Select all

cmartctl -d ata -t long /dev/sda

to start off a full test (results will be show in the full report, or see the manual pages for how to check progress and view only test results).

For testing performance you could use "hdparm -tT /dev/sda" to get a ballpark figure. A proper testing tool like bonnie++ will give you more realistic results though.

Rob1n, thank you for the prompt reply. What should I be comparing the numbers to? Basically, once I receive the results how do I know "yes my drive is under performing" or "everything is normal"?

[Rob1n]

Well, if you have two of them then I'd say comparing it to the other one should give you an idea

On a standard IDE drive I get:

/dev/hda:
Timing cached reads: 1718 MB in 2.00 seconds = 859.79 MB/sec
Timing buffered disk reads: 168 MB in 3.00 seconds = 55.95 MB/sec

The cached reads is affected more by memory and processor speed. The buffered disk reads is a more accurate measure of disk performance. For a standard SATA drive you should get about the same figure (they're nominally slightly faster but I don't think that really shows up), whereas for a SATA-II drive you'd be expecting a bit faster - in fact, I should have one of those hooked up around here somewhere. Ah:

/dev/sdl:
Timing cached reads: 1338 MB in 2.00 seconds = 669.28 MB/sec
Timing buffered disk reads: 204 MB in 3.03 seconds = 67.41 MB/sec

That system's a bit heavier loaded though.

[adsmith]

I had a new Seagate SATA drive last year which was giving similar error messages, which at the time I thought were due to hardware failure, but SMART logs were reasonably clean.

It turns out that the SATA drive or controller was not properly dealing with "unreadable block" error messages, and was spewing "OMG IM DYING" messages instead. It was also re-seeking a lot during this, which meant I heard the "click click click" which makes us all cringe.

Anyway, the error messages stopped after marking the badblocks on the filesystem. E.g., e2fsck -c -c /dev/sda1. All is well, just a bit of improper bad block handling in the hardware or driver.

[Carlo]

What should I be comparing the numbers to?

When the self detection notices something is really wrong, you've a value in the WHEN_FAILED column. The Problem is, that the chance is around 50% to have a dead harddisk, without getting such a notice before.

[klees]

Good point on comparing the two disks but do filesystems play a role? My first drive is my Windows XP drive which is NTFS (RO). Will there be a difference?

[Rob1n]

Good point on comparing the two disks but do filesystems play a role? My first drive is my Windows XP drive which is NTFS (RO). Will there be a difference?

Not in the hdparm results, no - it works on the raw disk (doing read-only tests) rather than on a mounted partition.

[klees]

Good point on comparing the two disks but do filesystems play a role? My first drive is my Windows XP drive which is NTFS (RO). Will there be a difference?

Not in the hdparm results, no - it works on the raw disk (doing read-only tests) rather than on a mounted partition.

Below are the results from hdparm:

Code: Select all

# hdparm -tT /dev/sda /dev/sdb

/dev/sda:
 Timing cached reads:   900 MB in  2.00 seconds = 449.99 MB/sec
 Timing buffered disk reads:  160 MB in  3.02 seconds =  52.97 MB/sec

/dev/sdb:
 Timing cached reads:   898 MB in  2.00 seconds = 448.35 MB/sec
 Timing buffered disk reads:  174 MB in  3.03 seconds =  57.49 MB/sec

Moreover, I ran a Samsung specific test on both drives and not a single error was found (this includes deep scanning). So now I'm wondering whether a recent update (maybe genkernel?) is causing the scsi_wait_scan message during boot as well as 'df -h' slowing down when reading the FAT32 partition. Anymore ideas?

[klees]

Low and behold, it seems to be related to a latest update and I believe the latest genkernel update is introducing these issues. The reason for this is, I start on an older kernel 2.6.21.6 and I get no scsi_wait_scan plus no slowdowns on any commands. I also see that the scsi_wait_scan module is not loaded on the 2.6.21.6 kernel whereas in the 2.6.22.1 (built recently) it does load. However, I do not have the SCSI_SCAN_ASYNC option enabled in the kernel yet scsi_wait_scan module loads in the 2.6.22.1 kernel. Does this make sense (I'm still a noob)?

I notice any kernel I built with the latest genkernel has scsi_wait_scan module loaded at boot. Is this a bug?

[klees]

Ideas?