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."
Old news is old (Score:5, Informative)
Yes, already saw some crazed guy at Sun shout into an array of HDDs and it decreased performance.
http://www.youtube.com/watch?v=tDacjrSCeq4
I saw the presentation... (Score:5, Informative)
..at SustainIT 2010, Turner had a really good analysis. Still some gaps - figuring out what frequencies hurt the most, and how individual drive types respond to what, is necessary followon. How various vendors' drive units transfer vibration from the rack into the unit, into the drive carrier, into the drive. That sort of thing. Now that the phenomena is identified, a lot more to do on it.
At the least, keep performance sensitive drives away from large sources of environmental vibration, such as your AC unit and so forth.
Re:Anyone know why it affects SSD? (Score:5, Informative)
the article does not say that it affects SSDs, but that it affects the SSD value proposition (aka, if you can spend little $$$ on carbon rack enclosures and get a significant seek performance increase, spending the large amount of $$$ to go full-ssd might not be as cost effective).
Re:Can't believe it (Score:3, Informative)
Some of the largest consumers of hard disks are enterprise companies with network attached storage, storage area networks, and RAID equipped servers.
These companies have a very high density of hard disks, and spend a lot of money for high performance. It's not unheard of for such a company to purchase a huge array, not for capacity purposes, but for seek time and throughput.
I'm sure they care.
Re:not surprising really (Score:3, Informative)
should be possible to add a sensor (probably even on silicon) that can warn of higher vibration and slow the stepping, but I bet they don't do that right now.
Re:not surprising really (Score:4, Informative)
I just want to point out that hard drives stopped using stepper motors [wikipedia.org] decades ago. They've used voice coils [wikipedia.org] since, which is basically an electromagnet and strong magnet which it deflects to various positions based on the field strength; in other words, it's continuous, not discrete like a stepper motor (though they can do microstepping [wikipedia.org] as well). OK, so in a way, a voice coil is sort of like a stepper motor with only one phase, which is then microstepped...
Well you are half right, but so is the GP.
The GP's description is more accurate if you replace "stepping" with "accelerating".
The head does not move to a position based on field strength (open loop control). It is free to move on low friction bearings, the applied field strength accelerates the head. Closed loop control is needed to make it stop at the correct position.
Re:Hmmm... (Score:4, Informative)
A perfect example of why this type of joke flies right over the average /.er's head.
So perfect, indeed...
Re:Can't believe it (Score:4, Informative)
unless you really stacked the deck against steel racks - loose screws, hard drives not properly mounted...?
Depending on where exactly they are, loose screws might actually help you. Tight metal-to-metal connections are much better at transferring vibrations, especially the higher frequency ones, than looser connections, where some of the vibrational energy is converted into lower-frequency vibrations. Steel is insanely good at carrying vibrations over long distances, hence the old movie trick of listening to railroad tracks for a train in the distance, or tapping on pipes in Morse code to communicate your escape plans to the inmate several cells over. (At the risk of veering off-topic, neither of these tricks work nearly as well in real life as in the movies, but they do work. Well, at least the railroad tracks do. Since the MPAA hasn't found my gargantuan mp3 collection yet, I haven't had a chance to test prison telegraphy yet.)
One thing that has always baffled me is why racks and computer cases are made of metal to begin with. There are, of course, certain areas where you need steel or aluminum for strength or carrying waste heat, but wood or plastic would do a much better job of damping vibrations. There's a reason audio speaker cabinets are made out of crappy, soft stuff like particle board: you don't want the cabinet to resonate, and particle board does a wonderfully poor job of transmitting vibrations, which is why it isn't used in guitars, where you want strong resonance. There are also a wide variety of synthetic rubbers like neoprene and sorbothane that do a good job of absorbing vibrations. Neoprene is cheap, and sorbothane, while more expensive, is still affordable and does such a good job of deadening vibration that it feels remarkably like meat. (I happen to have a square foot of it sitting on the counter next to me, waiting to be used in some vibration-damping experiments with my scooter, but having RTFA, I think I'll try using a little bit to replace the rubber pads on the bottom of my external drive enclosures.)
Re:Can't believe it (Score:3, Informative)
One thing that has always baffled me is why racks and computer cases are made of metal to begin with. There are, of course, certain areas where you need steel or aluminum for strength or carrying waste heat, but wood or plastic would do a much better job of damping vibrations.
It's for RF protection in both directions. Those acrylic cases can cause problems with the PC, or with wacky noise in your stereo. I'd stick with metal, but float the drive cage, and float each drive in the cage.
Re:Old news is old (Score:1, Informative)
Re:not surprising really (Score:2, Informative)
The head does not move to a position based on field strength (open loop control). It is free to move on low friction bearings, the applied field strength accelerates the head. Closed loop control is needed to make it stop at the correct position.
The servo firmware engineers I worked with went to great lengths to define and maintain (in real-time) the acceleration profiles for the actuator arm. The point is that the seek algorithm is fully closed loop.
Originally they walked the fine line between all-out performance and reliability, but later they started slowing down the seeks (on 7200 RPM drives) to make them quieter.
Re:not surprising really (Score:2, Informative)
The same techniques that silent PC nerds use to isolate their hard drive to keep it quiet would of course help reduce vibrations (I mean, that's the point). There is the thread detailing the techniques used to suspend/isolate HDD's at http://www.silentpcreview.com/forums/viewtopic.php?t=8240/ [silentpcreview.com].
Re:not surprising really (Score:2, Informative)
I used to work in the HDD industry, and am very familiar with the seek algorithms used. Sorry, but your description of the move algorithm is completely wrong. Modern control implementations collect information about the actual output of the system (such as position) relative to the desired output (the target position) and act on it in real-time.
Modern HDDs (as in, anytime in the last 15+ years) have 'servo tracks' written on the disk. These are radial spokes of distance information encoded on the disk. Today's drives may have as many as 500 or 1000 of these servo spokes. The head is able to read these as it seeks across the disk, so it knows where it is during the seek. There is no 'stopping, checking where you are and making an adjustment.' It's more along the lines of 'checking your position as you move, and adjusting the voice coil voltage to give the optimum velocity and acceleration to land in the right spot as quickly as possible.'
There is an enormous amount of effort that goes into reducing the amount of vibration generated by the drives themselves, as well as minimizing the drives' sensitivity to external vibrations. However, there are specs for what the drives are able to handle, worst-case. By reducing the amount of vibrations they are subjected to, they will naturally perform better.
Being Slashdot, we need a car analogy, right? Cars are made to deal with headwinds and still get acceptable gas mileage, but removing the headwind will increase a car's gas mileage.