Vibration Killing Enterprise Disk Performance? 159
An anonymous reader writes "Is vibration killing disk performance? ZDnet reports on research that a carbon fiber anti-vibration rack increased random read performance by 56% to 246% and random write [performance] by 34% to 88%. Vibration is a known disk problem, but this is one of the few attempts to quantify its impact — which looks to be much greater than suspected."
not surprising really (Score:5, Interesting)
It's a lot easier to interfere with a moving head arm than it is to mess one up that's locked on a track, so this isn't surprising in the least for vibration to affect reads that require numerous long seeks. I'm surprised it's not worse than they've found.
Moving the head requires accelerated head stepping to top speed, stepping to close to the track, slowing down, stopping at the destination track, waiting for the head to settle, and reading an address block to find out where you managed to land. If you find you missed the track, you have to go through the whole seek process again. (usually only once more, those short adjustment hops are pretty reliable because they're lower speed) But that really hurts your single block read time.
Add to that the fact that the "high performance" drives are making more risky higher speed track changes, which increase the odds of missing your target and make the operation more sensitive to vibration. I've written direct HDD io code before, and sure, you can up the step speed to get very nice seek time boosts, but then you start missing your track and start getting reseeks. Usually you go with the fastest that's acceptably reliable, and that puts you on the bleeding edge of having problems, where things like vibration can run you off the deep end of the bell curve.
It wouldn't surprise me one bit if 50% of the "high performance drive" better speed is due to faster spindle speed, and the other half is faster (riskier) seek speed.
Re:Can't believe it (Score:5, Interesting)
I've done some quick and dirty followups. Drive arrays on a concrete floor are much faster than those in a normal steel or aluminum rack with more drive arrays.
This is real.
You can demonstrate it with one drive array, a rack, and a solid floor. Put other equipment in the rack. Put the array in, test it. Stop testing, put it on the floor, start testing again. Put it back in the rack, test it again. The floor, test it again.
There are some time delays involved as the drives adapt to higher and lower vibration environment - the mechanism here is the drive seems to be adopting a strategy of more error correction on reads and writes when it thinks the head's vibrating more. It will ramp that up and down as it figures out that the environment has changed.
This is what you get... (Score:5, Interesting)
Re:not surprising really (Score:5, Interesting)
For the computers in my music project studio, I've always used Scythe stabilizers for my hard drives, mostly to keep the sound down.
But now that I think about it, the drives in my raid box are like 4 years old and not one has failed. I've thought about buying new drives to replace them, but as long as they're working and everything is backed up, I haven't bothered.
I don't know if the stabilizers have had any effect on their longevity, but the little bit of care I take with vibration dampening when I build machines sure has resulted in some quiet machines.
shouting at disks (Score:4, Interesting)
So maybe sex in the server room IS a good idea! Not to mention all the puns you can make about racks.
Also, don't shout at your disks since it increases latency, as the guys at Sun Storage learned:
http://www.youtube.com/watch?v=tDacjrSCeq4
This could be just a matter of resonance (Score:4, Interesting)
The disks all spin at roughly the same frequency (250Hz for a 15K RPM drive), so you could get some interesting resonance patterns in that frequency band as well as in its harmonics and frequencies that you get when you subtract rotational vibration spectrum of one drive from another. You can even hear these effects if you run two 7200 RPM drives in your desktop in a quiet room (assuming you don't have a dozen fans in the case that some people like to have for some reason).
The solutions is simple - dampen the drives to eliminate high frequency vibration transfer. Better yet, don't use screws to attach your drives at all. Use velcro.
SSD (Score:3, Interesting)
Re:Can't believe it (Score:5, Interesting)
My personal anecdote is: Working for a small company dealing in terabytes of data (7 years ago), they got their first disk array. Previous to that, they were using desktops to store everything around the network. So, after months of pleading, they got me the disk array I wanted, and the failure rate was atrocious. Averaged to 1 disk per 90 days. The SAN we used sat on a flimsy filing cabinet right next to a high speed printer. Not touching, but close. After a while of trying to figure out the problem, I finally sold the bosses on the idea of turning one of the closets into a server room. I installed a rack, mounted it to the wall with dampeners, and installed the SAN into it. Along with 2 1au servers, and another brand new NAS. The failure rate plummeted. The original SAN so prone to killing disks worked it's ass off for 2 more years before any of the drives failed again. As far as I know it's only had 3 disks replaced in the 5 years since then. Seems reasonable to me to assume that vibration not only plays a role in performance, but in lifetime as well.
My CDROM killed 2 hard drives (Score:2, Interesting)
Couple of comments from a storage guy... (Score:5, Interesting)
Note: I'm not a server guy, i'm not a LAMP guy, I'm not a OS guy nor a DBA. I deal with two objects in the datacenter, storage devices and switches that move packets (either FC or FICON) to storage devices.
First when I heard the term 'enterprise disks' I figured they were talking not about the drives themselves, but rather of the highend EMC, IBM, HDS variety (HP/SUN don't count b/c they rebrand HDS). There's no dampening in those arrays, they're basically racks/cabinets on wheels with casters on them to lock them down. If I was reading from a raw disk, then I could definitely see how vibration would have an impact, but with Enterprise disk arrays, there's so much cache (in the array, not on the drive) and read-ahead algorithms in place that I could see how users wouldn't notice the difference. I'm not so sure that EMC/HDS/IBM would be willing to build their disk arrays out of carbon fiber. Especially with the price conscious consumers like myself that love nothing more than my yearly meeting with my storage vendor to discuss $/GB.
I know of some companies that put their highend (Superdome/p690) servers on earthquake pads, which in the event of an earthquake the server can stay put while the floor shifts underneath.
I've actually experienced this problem first hand. I used to work at one of the above mentioned storage companies and we manufactured a disk shelf that had 8 drives in the front and 8 in the back. There was a metal divider in both the front and back that separated the box into quadrants. We noticed one year that there was a significant drive loss in the field and upon further investigation, we noticed that one slot in particular had an abnormally high failure rate. So we flew to one such site that had these suspect drive shelves, and measured the vibration of each disk in their disk shelves (they had about 100) using a tool that look pretty much like an accelerometer at the end of a pencil. Turns out that the drive location that had the highest number of failures, was not abnormally vibrating, but that a drive 4 spots away was. It seems that if a the drive next to the divider had a "high vibrational drive" it would set up a standing wave which would eventually cause another drive, which was perfectly fine from a manufacturing standpoint to fail.
Not surprising... (Score:3, Interesting)
Luckily, all my drives are mounted through silicone vibration isolators. Gotta love Antec case design! :D http://media.giantpachinkomachineofdoom.com/2007-02/blog/images/u1_array_supersize.jpg [giantpachi...ofdoom.com]