New drive diagnostics

[danwood82]

Hi all,
I've got an 8-disk raid array to set up, and I want to put the brand new drives through a thorough test in the hope of catching any catchable issues early.

Only problem is, I really don't know much about hard drive testing. What's a good process for vetting new drives and flagging up any issues.

I had heard it was a good idea to just grab the manufacturer's utility, so I downloaded SeaTools, and it appears to be a bit on the basic/half-finished side. It's not really clear what tests are worth running and which aren't either.

I've also read that just a non-quick full drive format will do a perfectly good job of mapping out bad sectors.

What are people's experiences?

 

[wyluliraven]

SMART data is another decent place to check out, CrystalDiskInfo seems to be a popular software.

Other options are here, however their first recommendation is Crystal.

 

[danwood82]

Cheers. Yeah, I'd come across a few of those options searching around.

The problem I'm finding is that there's plenty of suggestions for which bit of software to run, many of which allow you to check a million and one things, and run just about any test configuration you care to run. There seems the be very little advice on a good, solid process of what to check for and how.

Does anyone have a tried-and-tested method for putting their newly-bought hard drives through their paces before putting them to work?

 

[wyluliraven]

/snark filter installed

You could you know... try using them.

 

[danwood82]

/snark filter installed
You could you know... try using them.

I have every intention of using them. I've got 8 drives, that's 8-times the likelihood of a manufacturing defect, and as I intend to use them in a fairly heavy workload environment, I would prefer to discover any reasonably easy-to-catch errors before they bite me in the ass a few weeks down the line. It will be much simpler to send a drive back for replacement now, before I've even built the array, than to start running round like a headless chicken when I'm neck-deep in a project and need to replace a dropped disk that I could have potentially caught sooner.

I know I'm not going to ensure perfect operation, but as I have a few days until other equipment turns up for my server build, I think it would be time well spent to set them running some thorough diagnostics. I imagine that is more likely to show up any potential errors than just... using them.


I'll run a few tests regardless of advice, I was just reaching out to see if anyone had more experience and had a bit more methodical process for giving a drive a once-over, which would improve on my "run that test because it sounds good" approach.

 

[wyluliraven]

SMART data is exactly what most people use to identify drives with issues, besides being DOA.

Can't really continue in good conscious continue to reply any more because you seem to rebuke and casually disregard the advice you solicited and call it hokum. Especially given your earlier statement about Seagate's own software.

so I downloaded SeaTools, and it appears to be a bit on the basic/half-finished side.

 

[danwood82]

SMART data is exactly what most people use to identify drives with issues, besides being DOA.
Can't really continue in good conscious continue to reply any more because you seem to rebuke and casually disregard the advice you solicited and call it hokum. Especially given your earlier statement about Seagate's own software.

I'm not rebuking anything, or calling anything hokum. There hasn't really been anything to rebuke. I thanked you for the suggestion, and then elaborated a little more on what I was seeking, as I don't suppose it was quite clear enough in my original post. I was just clarifying that I was more inquiring to see what processes people had used to test drives out, rather than just suggestions for which software to run.

I'm aware that SMART data is a good thing to check, but as far as I'm aware, SMART doesn't analyse the actual state of the drive, it just reports/logs errors and glitches as they occur.

Regarding Seagate's own software - it asks me to "Agree/Disagree" to a blank page on startup, the tests have rather vague names, if I click on "help" to get more information, it just throws an error, and when I run a test, it gives me very little indication that it's even doing anything. It seems a little basic/half-finished to me. It may very well do it's job, I'm running some tests now, but I'd like a little more indication of what is advisable to run, and why.

 

[PigLover]

Danwood's question is not unreasonable - especially when running consumer drives behind a hardware raid controller which might hiccup if the drive has a questionable spot on it and takes too long trying to do its own error correction.

SMART won't tell you anything on a brand new from the manufacturer drive. It is not predictive. It only records things that happened during operation (like pending and reallocated sectors). Until you light up the drive and use it for a while the SMART data will show a perfect drive.

Good options are to load up a Linux image and run badblocks. Alternatively, get a copy of the Ultimate Boot CD and run the windows surface analysis tool.

After you run a surface analysis tool - even if badblocks or WSAT didn't report any errors. After running the scan SMART data will have recorded anything the drive "fixed" that the scan reported as a good result.

 

[danwood82]

Danwood's question is not unreasonable - especially when running consumer drives behind a hardware raid controller which might hiccup if the drive has a questionable spot on it and takes too long trying to do its own error correction.

SMART won't tell you anything on a brand new from the manufacturer drive. It is not predictive. It only records things that happened during operation (like pending and reallocated sectors). Until you light up the drive and use it for a while the SMART data will show a perfect drive.

Good options are to load up a Linux image and run badblocks. Alternatively, get a copy of the Ultimate Boot CD and run the windows surface analysis tool.

After you run a surface analysis tool - even if badblocks or WSAT didn't report any errors. After running the scan SMART data will have recorded anything the drive "fixed" that the scan reported as a good result.

Thanks PigLover.
I'm hoping the drives play ball with a RAID controller - they're Constellation CS drives, which I know are basically consumer drives, but I think they're at least firmware-tweaked for enterprise environments.

badblocks sounds like a good one to try out, I assume I could run that easily enough from a Linux Live CD... I might give that a shot.

Do you happen to know a link to the standalone application for "windows surface analysis tool"? It sounds like a standard Windows component, but I'm not aware of it.
I'm testing the un-partitioned drives from a working machine, so there's no particular need for me to dip into DOS.
I've tried searching, but I only turn up links to the damned "Microsoft Surface" Can't see any specific reference to it when looking at the tool list on Ultimate Boot CD - Overview

 

[PigLover]

Badblocks should be fine off a live-CD.

Sorry - brain fart - Windows Surface Scanner isn't on the current UBCD (licensing issues...).

As long as you are comfortable with Linux and running off a live-CD then badblocks should do what you want.

 

[RimBlock]

Thanks PigLover.
I'm hoping the drives play ball with a RAID controller - they're Constellation CS drives, which I know are basically consumer drives, but I think they're at least firmware-tweaked for enterprise environments.

Would love to hear your views on the CS drives as there seem to be no benchmarks or anything other than "Seagate announced..." on these drives. They appear to be good on paper and are pretty well priced. I am considering getting a few myself so any feedback on them would be great.

RB

 

[Chuckleb]

I too would be interested to hear how the Constellation CS run. We've been buying WD Red drives (WD30EFRX) and they seem to be pretty stable. We're running them a bit larger than their recommended NAS configuration, but they don't drop or have any problems. Plus they are cheaper than the Seagate CS. The run fine in both software raid and as part of LSI volumes.

 

[dba]

I have a rather simple test and burn-in procedure for drives:

1) Mount the drive in a Windows box (because IOMeter runs on Windows).
2) Format the drive. Full format (slow) for non-SSD drives and quick format for SSD drives.
3) Spin up IOMeter and beat the heck out of the drive overnight. For SSD drives I test only reads. For non-SSD drives I do a mix of reads and writes over the entire drive, usually 80% reads and 20% writes.
4) In the morning, check the IOMeter UI for errors. There must be none.
5) Use a tool (e.g. speedfan) to take a peek at SMART for errors and issues.

The above seems to catch any drives that are prone to early death. At one point, for example, about 5-8% of OCZ Vertex3 drives would fail while the others would run forever with no issues.

Next, I test each drive to make sure that it meets expectations:

1) I use IOMeter to test 1MB random reads because that is my primary use case, but testing reads and writes of various sizes with a tool like Anvil might be more general

The idea behind this second round of testing is to confirm that each drive performs just about the same as its peers. I do find that a small percentage of my new drives are, for some reason, quite a bit slower than they should be. I send these back right away.

This might seen like overkill, but if you add a new drive to an eight-drive RAID6 array and it fails, you could end up spending days getting back to stable.

Hi all,
I've got an 8-disk raid array to set up, and I want to put the brand new drives through a thorough test in the hope of catching any catchable issues early.

Only problem is, I really don't know much about hard drive testing. What's a good process for vetting new drives and flagging up any issues.

I had heard it was a good idea to just grab the manufacturer's utility, so I downloaded SeaTools, and it appears to be a bit on the basic/half-finished side. It's not really clear what tests are worth running and which aren't either.

I've also read that just a non-quick full drive format will do a perfectly good job of mapping out bad sectors.

What are people's experiences?

 

[cafcwest]

posting so I can find this later.

 

[danwood82]

Thanks dba, that sounds like it could be a good process. I might give that a try.
I've been trying out HDDScan, and while its surface scan's block-access-latency information is quite interesting, it might be a better use of time to actually stress-test things a little.

@Chuckleb @RimBlock
- I'll be sure to post any findings on here. So far, they seem like good, solid, ordinary drives. Very much like they appear - a tweaked version of the 7200.14. Same read/write speeds and latency, but rated for 24/7 use instead of the 8-hours-per-day rating for the desktop drives.

I'll be building these into a dual Xeon E5-2687W hybrid-server-workstation, just as soon as scan.co.uk get stock of the S2600COE motherboard I've ordered.
I'll crack out the camera when I come to do the full build, and hopefully post that up here too.

(PS, I know I should have a separate file server, but this is a special-case... I'm using the array primarily as an immense scratch-disk for fluid sim data. It needs to be very fast and very very big, so it makes more sense to connect it direct)