Why are there no MTBF data for sbc's

Moderators: mdrjr, odroid

Why are there no MTBF data for sbc's

Unread postby back2future » Fri May 11, 2018 2:50 am

Why does not Hardkernel support information about mean availability of hardware before one could expect loss of function because of wearout/workout?
Thanks, rough estimation would be precise enough for us.
Last edited by back2future on Fri May 11, 2018 9:15 am, edited 1 time in total.
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Why are there no MTBF data for sbc's

Unread postby odroid » Fri May 11, 2018 4:02 am

Because we don't know how to generate the MTBF data, we could not provide it.
But we can guess the eMMC or SD card wear out time seems to be shortest one.
If your application has very frequent write access to the flash storage, you have to consider the total number of write cycles of the NAND flash based storage.
Refer the reliability data on this MLC type eMMC datasheet page 18.
http://www.mouser.com/ds/2/669/SanDisk_ ... 128853.pdf
User avatar
odroid
Site Admin
 
Posts: 28490
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: Why are there no MTBF data for sbc's

Unread postby back2future » Fri May 11, 2018 9:14 am

believe it or not _ one method for MTBF, but not a scientific one :idea:
https://en.wikipedia.org/wiki/Centennial_Light

How many power-on hours have been aggregated by the longest used (production) sbc's at Your company?
Probably s5pc100 soc started at 2009? Repairable parts?
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Why are there no MTBF data for sbc's

Unread postby mad_ady » Fri May 11, 2018 3:26 pm

It will heavily depend on usage patterns - a mining sbc will age faster than an idle server sbc. I'm relatively new to odroids, but my oldest unit is still running after ~2 years
User avatar
mad_ady
 
Posts: 4214
Joined: Wed Jul 15, 2015 5:00 pm
Location: Bucharest, Romania
languages_spoken: english
ODROIDs: XU4, C1+, C2, N1

Re: Why are there no MTBF data for sbc's

Unread postby rooted » Fri May 11, 2018 3:27 pm

Meveric has some U2/3's he has worked hard for years.

I have an XU3 and C1 that still work fine, had since release. I also have two XU4's in 24/7 operation going on a couple of years now.
Last edited by rooted on Sat May 12, 2018 12:04 am, edited 1 time in total.
User avatar
rooted
 
Posts: 5297
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1, C1+, C2
XU3 Lite, XU4
N1
VU7+
HiFi Shield 2
Smart Power (original)

Re: Why are there no MTBF data for sbc's

Unread postby back2future » Fri May 11, 2018 8:07 pm

Just curious about replacement costs.
Like for harddisks one could assume averaged design life time at ~5 years¹.
( just mentioning os compatibility and specified application-related data for that endurance also )
Xu4 for 5 years would demand ~15$ for each year of duty (while energy costs for 24/7 are probably above 10$ on US electricity supply).
Additional heat spreader could provide lower base temperatures (less important (within specification range) for hdd during mean life time, more important for older hdd for example).

Overclocking is lowering mean availability of cpu parts, while underclocking is supportive for enhanced life time expectance of cpu parts (maybe that's comparable for external gpu, not soc included ones and ram parts). Interesting details are points, where we significantly leave linearity.

Additional vibration testing?
[ https://magazine.odroid.com/article/and ... roid-road/ ]

[ 1) https://www.backblaze.com/blog/hard-dri ... s-for-2017 regarding current numbers of harddisks installed compared to harddisks replaced 2013-2017 there were 1/4 after 4year 24/7 anymore functional inside a datacenter situation, so a often read 5year guess could be a suitable averaged assumption for consumer hdd ]
Last edited by back2future on Sun May 13, 2018 12:03 am, edited 3 times in total.
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Why are there no MTBF data for sbc's

Unread postby back2future » Sat May 12, 2018 9:27 pm

odroid wrote:Because we don't know how to generate the MTBF data, we could not provide it.
But we can guess the eMMC or SD card wear out time seems to be shortest one.
If your application has very frequent write access to the flash storage, you have to consider the total number of write cycles of the NAND flash based storage.
Refer the reliability data on this MLC type eMMC datasheet page 18.
http://www.mouser.com/ds/2/669/SanDisk_ ... 128853.pdf


MTBF for cooling fans (ball/sleeve bearing comparison: failed 5%/45% after 5yrs@140°F/60°C)
https://www.sepa-europe.com/en/technisc ... tancy-mtbf

Regarding, Reliability of iNand emmc HS400, page 18, it would be useful having storage sizes explained as available to write, before having to move written cells data (like shown by df cmd), for Write Endurance Specification (TBW)? Is it necessary having (sometimes recommended) 20% free disk storage, because of wear leveling, on HK emmcs (at least after a 0.25k/10% cells write access @55°C/131°F or older than 3-5yrs)?
(https://en.wikipedia.org/wiki/Write_amplification)
Some file systems that rotate data because of handling data like sequential written circular logs are JFFS2, YAFFS, UDF, ZFS, UBIFS or NOVA.

'Cycles, Cells and Platters: An Empirical Analysis
of Hardware Failures on a Million Consumer PCs'
2011, Microsoft research, Conclusion, page 12 and figure 15, page 13:
https://www.microsoft.com/en-us/researc ... ingale.pdf

Companies tell:
one rare exception to this topic: 5yrs product guarantee: https://www.versalogic.com/applications ... ns.asp?m=1
Lifetime cycles from 5-7-10 yrs are more frequently advertised and guaranteed, but that's only a hint for duration of innovation periods on this subject.

Reliability engineering:
https://en.wikipedia.org/wiki/Reliability_engineering
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Why are there no MTBF data for sbc's

Unread postby ASword » Fri Jun 29, 2018 9:06 am

MTBF data ignoring flash lifespan would be useful — i.e. how long with the SBC typically last, if you replace the flash devices as necessary? Each component in the design ought to have an MTBF figure available for it, and there are standard approaches for combining this data to calculate an aggregate figure.

I’ve heard that short life capacitors are usually the culprit in PCs. Would be good to know what it is in HK’s designs, and perhaps new designs (N-2?) could avoid choosing parts with particularly limited lifespans.
ASword
 
Posts: 76
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: Why are there no MTBF data for sbc's

Unread postby back2future » Fri Jun 29, 2018 5:16 pm

Most frequent issues here with sd-cards (production date ~10 years ago) loosing their default capacities with usage at almost rated storage capacity. Cards showing wrong storage sizes (3.81GB became 2.81GB or 6.xxGB) when changing file system format. ( Not much experience around with uncommon, but flash optimized file systems, for e.g. f2fs and for Linux and Windows systems. )
Udf format with ImageWriter software to usb flash made these read-only.
One hard disk started producing power issues after ~18000h, but data were still recoverable.
One usb 3.0 thumb nail usb flash storage always overheated, reduced write speed then and finally failed.
Modern ssd can have multi core storage controller only for wear levelling optimization.

There are issues to call, but (usb mounted) storage for customers is easy to replace.

Most parts on a pcb will benefit from cooler temperatures, especially (electrolytic) capacitors, afaik.
One of xu4 has an additional mounted fan that keeps overall temps mainly below 35-45°C (values differ with kernel 3.10 or 4.x) on idling.
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english

Re: Why are there no MTBF data for sbc's

Unread postby ASword » Thu Jul 12, 2018 3:14 am

The flash lifetimes can be managed separately from the rest of the device (especially for removable/replaceable media). As for cooling, I plan on using a passive device with large heatsink in combination with large diameter temperature controlled fans that cover the larger overall system.
ASword
 
Posts: 76
Joined: Fri Aug 04, 2017 12:48 pm
languages_spoken: english
ODROIDs: XU4, HC1

Re: Why are there no MTBF data for sbc's

Unread postby back2future » Fri Jul 13, 2018 9:55 pm

There is some discussion on actively cooling peripheral connectors that are at Gbit speeds or for high power transfer.

A TFBGA package for a rk3399 ( junction_ambient thermal resistance: theta ~12°C/W ) could require ~40-160in2 case area for dissipating ~10W to low velocity air, ambient temperatures being at ~70-105°F.

[ 1) https://www.engineeringtoolbox.com/conv ... d_430.html ]
naturally beYOnd spectrum
back2future
 
Posts: 75
Joined: Sun Jul 23, 2017 3:19 pm
languages_spoken: english


Return to Hardware and peripherals

Who is online

Users browsing this forum: No registered users and 2 guests