[AusNOG] [AUSNOG] Disk wear & Foucault Period

Matthew Moyle-Croft mmc at mmc.com.au
Thu Aug 22 12:21:55 EST 2019


I’m not aware of any statistically significant data showing that latitude or direction of HD has an affect on disk life time nor direction of the racks even from companies with *extremely large* global fleets.

I also think before making assertions as per below that you’d need to demonstrate that this force is in anyway actually significant. There are lots of forces and physical effects acting on equipment. Vibration, temperature, actual usage patterns. You’d also need to understand the mechanics of the drive - does this force actually mean that the bearings move in any meaningful way?

eg.Vibration effects https://www.usenix.org/legacy/event/sustainit10/tech/slides/turner.pdf
eg.Different cooling strategies: https://www.usenix.org/system/files/conference/fast16/fast16-papers-manousakis.pdf

Have you calculated the actual forces here and compared them to others? I don’t doubt precession exists, but I doubt it has any measurable affect on the MTBF of HDs. 

MMC


> On 22 Aug 2019, at 10:39 am, Paul Wilkins <paulwilkins369 at gmail.com> wrote:
> 
> Btw, if the forces seem insignificant, consider that turning a disk over 360 degrees every day, means precession forces doing work equivalent to 2 full start/stop/start cycles every day * sin(latitude). It's worse than this of course because of the shear forces on the bearings.
> 
> Over a disk's 5 year life, that's about 1800 power cycles. Not enough to kill a disk, or to even be an obvious problem, but a hidden and unnecessary drain on disk life and IT budgets.
> 
> So yes, I'm now of the view we as an industry should insist that rows in our DCs should be run North/South.
> 
> Kind regards
> 
> Paul Wilkins
> 
> On Thu, 22 Aug 2019 at 10:17, Paul Wilkins <paulwilkins369 at gmail.com <mailto:paulwilkins369 at gmail.com>> wrote:
> I think this is germane to the mail list for the following reasons:
> 
> 1 - IF there is an obvious correlation between Au resident DCs with East West running cabinet rows and higher failure rates rather than North South running cabinet rows, then it should be within the ability/resources of this mail list to identify.
> 
> Reason being, for precession purposes, Australia qualifies as Equitorial (as opposed to Polar). The means you minimise precession with vertical disks, where the disk axis points north. Because disks insert into disk arrays face first, this means you minimise precession with cabinet rows that run North/South. (ie. the cabinet rows are parallel to the disk axes).
> 
> 2 - IF the correlation is real, then this is knowledge of value, due to improved reliability and level of service that entails. And just as importantly, tech time spent restoring crashed drives can be invested elsewhere. Thirdly, it may mean you get to extend the useful life of drive arrays, which will give capital and operational economies.
> 
> If there's no obvious correlation, the only cost is some argument over noise on the list.
> 
> Kind regards
> 
> Paul Wilkins
> 
> On Thu, 22 Aug 2019 at 08:45, andrew khoo <andrew.khoo at as136019.net <mailto:andrew.khoo at as136019.net>> wrote:
> i hate to be a “me too”.
> 
> i personally believe this issue is of relevance to operators.
> 
> in the australian context this is even more relevant due to potential costs we have to incur to keep spinning rust for our compliance requirements.
> 
> maintaining a healthy lifecycle means potentially extending the useful life of storage that surely an operator cannot run without.
> 
> just my 2c.
> 
> (and OP’s theory might mean a migration to DCs in darwin? :) :))
> 
> 
> 
> On Thu, 22 Aug 2019 at 08:37, <andy at coastalaudio.com.au <mailto:andy at coastalaudio.com.au>> wrote:
> Excuse my apparent naivety, but I thought data centres were attached to networks? It’s apparent that only DevOps engineers are allowed to think scientifically, unlike the rest of us mere mortals. I’m sure that I’m not the only one that found Paul’s theorising quite interesting…or has the inherent Luddite atmosphere of stifling creative thought in this country now permeated into the Technosphere…?
> 
>  
> 
>  
> 
>  
> 
> From: AusNOG <ausnog-bounces at lists.ausnog.net <mailto:ausnog-bounces at lists.ausnog.net>> On Behalf Of James Hodgkinson
> Sent: Wednesday, 21 August 2019 6:36 PM
> To: ausnog at lists.ausnog.net <mailto:ausnog at lists.ausnog.net>
> Subject: Re: [AusNOG] [AUSNOG] Disk wear & Foucault Period
> 
>  
> 
> Please stop thinking out loud, this is the quiet carriage. 
> 
>  
> 
> James
> 
>  
> 
> On Wed, 21 Aug 2019, at 17:17, Paul Wilkins wrote:
> 
> Another thought, which is that horizontal mounting is optimal for polar regions, whereas you minimise precession at equitorial latitudes with vertical mounting (but only if the axis is north aligned), which could go some way to explaining the anecdotal stuff you hear about horizontal versus vertical mounting. Though I've yet to hear anyone asking what's your latitude before they proceed to build a data centre or installing your vertical disks arrays to be north axis aligned.
> 
>  
> 
> Kind regards
> 
>  
> 
> Paul Wilkins
> 
>  
> 
> On Wed, 21 Aug 2019 at 16:31, Mark Smith <markzzzsmith at gmail.com <mailto:markzzzsmith at gmail.com>> wrote:
> 
> How is this related to network operation? AusNOG - Australian Network Operators Group.
> 
>  
> 
> You wouldn't go to a car show to ask about motorbikes - you'd to to a motorbike show, because that's where the most people interested in and having the most knowledge about motorbikes would be.
> 
>  
> 
> On Wed, 21 Aug 2019 at 15:30, Roy Adams <roy at racs.com.au <mailto:roy at racs.com.au>> wrote:
> 
> Paul, I would be most interested if you hear back from them.
> 
>  
> 
> I am curious if even BackBlaze have considered your idea in the past.. maybe a wake up for them and others
> 
>  
> 
> FYI, they produce these reports each quarter - worth calendaring to pick up.
> 
> I was stunned at the 14TB Toshiba reliability... interestingly, 2 weeks ago that top line read 0 failures... today it says 1 - v.strange
> 
>  
> 
>  
> 
> Kindly,
> 
>  
> 
> ROY ADAMS | P 07 3040 5010  | Web: http://www.racs.com.au/ <http://www.racs.com.au/> | Wiki: https://ex.racs.com.au:444/ <https://ex.racs.com.au:444/> | eMail: mailto:roy at racs.com.au <mailto:roy at racs.com.au>
> Please never upgrade to the latest Windows 10 - You don’t need the hassle, and I don’t need the work.
> 
> More seriously, the 6 month older Windows 10 releases are typically FAR MORE stable - a simple RACS script can fix this - just ask :)
> If you think it's expensive to hire a professional to do the job, wait until you hire an amateur - Red Adair.
> 
> Life is a journey through a series of adventures.. Live them, love them, hate them, but never give up on your dreams, desires, and goals.
> 
> Have you been good today? .ಠ_ಠ
> 
>  
> 
>  
> 
> On Wed, 21 Aug 2019 at 11:22, Paul Wilkins <paulwilkins369 at gmail.com <mailto:paulwilkins369 at gmail.com>> wrote:
> 
> Roy,
> 
> Those are the most comprehensive disk stats I've ever seen.
> 
>  
> 
> So out of curiosity, I emailed Backblaze to ask if they'd noticed any latitudinal effect on disk wear. Not realistically expecting a response, but if they did notice a correlation, it may perhaps emerge by osmosis.
> 
>  
> 
> Karl, I have to think turning a 15K RPM on it's head over a, admittedly long period, has to have an effect. Cumulatively you're going to get uneven wear on the bearings. Also the wear is going to be strongly non linear. Even if it's not a factor today due to current production tolerances and HVAC, it must become a factor eventually, though maybe not necessarily in our lifetime.
> 
>  
> 
> Kind regards
> 
>  
> 
> Paul Wilkins
> 
>  
> 
> On Tue, 20 Aug 2019 at 19:59, Peter Betyounan <peter.betyounan at serversaustralia.com.au <mailto:peter.betyounan at serversaustralia.com.au>> wrote:
> 
> Easy solution , move to all flash, predicable failures and wear. 
> 
>  
> 
> Tue, 20 Aug at 4:07 pm, <paulwilkins369 at gmail.com <mailto:paulwilkins369 at gmail.com>> wrote:
> 
>  
> 
>  
> 
>  
> 
> Has anyone ever noticed a pattern of disks in equatorial latitudes lasting significantly longer than say Sydney or Melbourne? I notice the Foucault Period for Brisbane is 52hrs vs Melbourne's 39hrs and can't but wonder if this doesn't mean Brisbane based DCs will have their disks last 30% longer, or Darwin x3.
> 
>  
> 
>  
> 
> Kind regards
> 
>  
> 
> Paul Wilkins
> 
> _______________________________________________
> 
> AusNOG mailing list
> 
> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
> http://lists.ausnog.net/mailman/listinfo/ausnog <http://lists.ausnog.net/mailman/listinfo/ausnog>
> _______________________________________________
> 
> AusNOG mailing list
> 
> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
> http://lists.ausnog.net/mailman/listinfo/ausnog <http://lists.ausnog.net/mailman/listinfo/ausnog>
> _______________________________________________
> 
> AusNOG mailing list
> 
> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
> http://lists.ausnog.net/mailman/listinfo/ausnog <http://lists.ausnog.net/mailman/listinfo/ausnog>
>  
> 
>  
> 
> 
>  <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient>	Virus-free. www.avg.com <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=emailclient> <x-msg://43/#m_-1607061083877202490_m_-830375872720839702_m_2296110433590984146_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>_______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
> http://lists.ausnog.net/mailman/listinfo/ausnog <http://lists.ausnog.net/mailman/listinfo/ausnog>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net <mailto:AusNOG at lists.ausnog.net>
> http://lists.ausnog.net/mailman/listinfo/ausnog <http://lists.ausnog.net/mailman/listinfo/ausnog>
> _______________________________________________
> AusNOG mailing list
> AusNOG at lists.ausnog.net
> http://lists.ausnog.net/mailman/listinfo/ausnog

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ausnog.net/pipermail/ausnog/attachments/20190822/495f8357/attachment.html>


More information about the AusNOG mailing list