How's your experience with SMR-harddrives?

Hi experts.

Although this SMR-technology http://en.wikipedia.org/wiki/Shingled_magnetic_recording is still quite new, I hope to find experts here, that have already used quite a number of those for quite some time, let's say opinions qualify if
->at least 10 for at least half a year.

I simply wonder how your experience with those has been - it's not about recommendations.
LVL 60
McKnifeAsked:
Who is Participating?
 
Gary CaseRetiredCommented:
The problem is most likely the delays when the persistent cache is in use ... the RAID controller sees this longer-than-normal delay as an error and simply fails the drive.    Same reason non-SMR drives need to have TLER settings or they'll fail also in RAID environments.
0
 
Gary CaseRetiredCommented:
I can't give you a "qualified" opinion ... I've used 3 of the 8TB Seagates  for about 3 months  (not close to your hope for 10 for 6 months) ==> but for what it's worth ...

These are called "Archive" drives -- and that's basically how I use them ... to store backups and a lot of essentially static media files; and they work VERY well with this usage pattern.

I have, however, tested them a good bit (when I first got them) to see just how bad the writes can get, since that's the major issue with the shingled technology.    Simple fact is they worked surprisingly well ... you really have to stress them to get them in the "ultra-slow" write mode that they'll hit if you get them to the point where there are a lot of shingled band rewrites queued.

Basically I would not hesitate to use these for data drives in a single user system; but would not recommend them for the main OS drive where there's a LOT of random disk I/O, or in multi-user setups (e.g. servers) where there is likely to be a lot more random write activity.

A couple key technical details I've learned from various research sites that explain why they perform so well despite the shingled technology:

=>  The drives have a "persistent cache" (25GB on the 8TB versions) ... a non-shingled area where any random writes are staged so the band rewrite that would otherwise be necessary is avoided [the sector is moved to where it belongs and the band rewrites done when the drive is otherwise idle].

=>  The drive's firmware recognizes if you're writing enough sequential data to actually fill a band [it's all held in the drive's memory buffer until it makes that determination] ... and, if so, it simply writes the full band (thus avoiding any need for a band rewrite).    So in this case, there's NO performance penalty relative to a standard SMR drive.    Any residual sectors in the write that won't fill the next band are staged in the persistent cache.

The result is that the drive works almost exactly like a standard SMR drive UNTIL the persistent cache is filled.    But that is actually very hard to do, since if you write a lot of large files most of them will probably be written directly to where they belong (since they'll fill full bands); and you have to write more than 25GB of data to the persistent cache to fill it.

When the drive DOES hit the "wall" caused by a full persistent cache and ongoing random writes that require band rewrites, the performance is abysmal ... it effectively "freezes" for a LONG time until it has done a lot of the pending band rewrites and cleared the persistent cache ... this can be an hour or more.    But except for a long (many hours) test where I wrote thousands of small files non-stop to random locations on the drive [Did random deletes followed by random replacements of those files] I have never encountered this situation in "real" usage of the system.

Bottom line:   I was VERY skeptical of the performance because of the restrictions of the shingled technology; but the simple fact is that Seagate has done an excellent job of mitigating those issues with their design and the drives work very well.   You wouldn't want to use them in an application where there's a LOT of random write activity from multiple users [i.e. an active database or a server where there are a lot of users, causing a lot more random distribution of the write activity] ... but for many applications they work just fine.    If your usage pattern is such that you don't write more than 25GB at once except when writing large files (which primarily bypass the need for the persistent cache anyway), you'll likely not even realize these are shingled drives.
0
 
McKnifeAuthor Commented:
Hi Gary.
Thanks for writing. My motivation for this thread is my experience with a handful (4) of those seagate archive drives as well which we hoped to use for archiving. They produce weird results (not performance related) to say the least, but I will not go into that.
It made me suspect this technology is not fully mature and I hoped to find opinions supporting/opposing to that thought. But it would be really important to have people talking about a little more than just one Raid and also long-time-experience.
I am aware that only one SMR model has been around long enough.
0
Ultimate Tool Kit for Technology Solution Provider

Broken down into practical pointers and step-by-step instructions, the IT Service Excellence Tool Kit delivers expert advice for technology solution providers. Get your free copy now.

 
Gary CaseRetiredCommented:
Seagate claims to have shipped over 1 million SMR drives ... but I've certainly not noticed their availability for very long, so I'm not sure who got the early shipments.   I assume the 5TB drives they were shipping last year were SMR, since they don't seem to have a standard PMR 5TB unit.

Not sure you'll find many folks with over 6 months experience with them.

Just out of curiosity, what are the issues you've seen?    Noise?  Vibration?   ... or actual performance?    I've certainly been happy with the 8TB units ... 1.33TB/platter areal density provides EXCELLENT read speeds (over 200MB/s on the outer cylinders) and I've had excellent write speeds as well.    Obviously with only 90 days or so of experience I can't comment on their reliability ... a 3 year warranty is nice; but clearly I'd prefer to never have to use it :-)
0
 
McKnifeAuthor Commented:
Let me answer when the manufacturer of the storage-system has made a move. A support case is still pending.
0
 
McKnifeAuthor Commented:
A few weeks later...
Those drives will never see our server rooms again. After copying for sometimes one hour, sometimes five, sometimes even 24 hours, the RAIDS (non-NAS) explode, the NAS server crashes (2 different NAS even) and so do our tempers.
These drives cannot be mature, sorry.

The manufacturer of those NAS' (QNAP) could not solve it, so we built our own LSI raid controller based storage system even - same story, crashes did go on. Exchanged several drives - no better. Case closed.
0
 
McKnifeAuthor Commented:
I've requested that this question be closed as follows:

Accepted answer: 0 points for McKnife's comment #a40838978

for the following reason:

See last comment
0
 
Gary CaseRetiredCommented:
You should have noted that you were thinking of using these in a RAID environment => clearly they are NOT designed for that at all.   I'd have made that point loud & clear !!     The shingled bands are a VERY POOR fit for any striping activity across drives.    I thought my outline of the way they worked implied that, but I never even considered that you'd be looking at them for use with a RAID controller.
0
 
McKnifeAuthor Commented:
There's no doubt that there are better drives for striping, but that's no reason to tilt the whole system. Alone four times the controllers said different drives are defective. These are bugs, no less.
I already had these drives, so no point in warning;)
0
 
McKnifeAuthor Commented:
Gary, I'd like to thank you for the info about the persistent cache - this might be it, we'll see soon.
0
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.