[OT] What are the odds...

Jan 07, 2013 10 Replies
Saved reading
Jump to replies (10)

Of a small Atom based fanless computer with a solid state disk that was designed to be bombproof, getting a bad sector


and that sector being the inode containing the directory data for /sbin/



?!!


It was a bit worrying[1] when on reboot it said "/sbin/init not found" (to non linux people, that's basically the program that boots everything else after the kernel has initialised. Having that directory missing is generally a Bad Thing (TM)


[1] As this is my router/firewall/DHCP server, I went apeshit.


Luckily, I had a backup copy of / on another partition (and now on my laptop) and managed to put /sbin back booting off a recovery USB stick.


I run mine off a CF card. And I have spares!

On Tuesday 08 January 2013 00:15 Bob Eager wrote in uk.d-i-y:

That's (almost) what I thought I was doing with an SSD - maybe they are more failure prone?...

I am running it close to the bone without fans - it's not a weedy processor, but it's in a case that is grid punched with holes over most of the top and sides. Knowing it *can* run without a fan, perhaps I should put a slow quiet one in just to drop the temps.

smartctl reported (for a long self test) "no errors" this morning...

There are onchip sensors to report chip temps on the CPU and generally motherboard as well.

Mine runs in a case with PSU with fan but that's all - no issues.

I suspect that the flash chip is more the issue than the CPU.

Judging by my general reading on the subject, all flash chips are not created equal., and death can happen when changing data - block erases are followed by individual writes to change even one sector.

I think flash will get better, but for now I wouldn't use it except as a temporary store of non critical data or in a read only manner.

The CF card is run pretty well read only. It's also in a slot on the front of the server so I can change it really quickly.

If an area of the SSD has failed I presume that it has been remapped to one of the very many spare areas included for just this eventuality. Presumably also done by the SSD firmware and concealed from the real world. So you shagged SSD is once more magically a virgin :-)

AFAICR this is one reason why SSDs are still best used as temporary storage devices for transient data (but not too transient) as various bits die and then are replaced by the large amount of spare memory tucked away in the wings. This doesn't stop people (myself included) using these as system discs but there are a lot of sad stories about people having SSDs fail if you look at reviews.

So - what are the odds? I think that if you run with a full system on an SSD for a number of years then there is a reasonable chance of filestore corruption at some level. However I am too lazy at the moment to do anything specific about this. Perhaps one should do an image backup and restore every 6 months?

Cheers

Dave R

...snip... If an area of the SSD has failed I presume that it has been remapped to one of the very many spare areas included for just this eventuality. Presumably also done by the SSD firmware and concealed from the real world. So you shagged SSD is once more magically a virgin :-)

AFAICR this is one reason why SSDs are still best used as temporary storage devices for transient data (but not too transient) as various bits die and then are replaced by the large amount of spare memory tucked away in the wings. This doesn't stop people (myself included) using these as system discs but there are a lot of sad stories about people having SSDs fail if you look at reviews.

So - what are the odds? I think that if you run with a full system on an SSD for a number of years then there is a reasonable chance of filestore corruption at some level. However I am too lazy at the moment to do anything specific about this. Perhaps one should do an image backup and restore every 6 months?

On Tuesday 08 January 2013 12:56 Paul D Smith wrote in uk.d-i-y:

What make of SSD would you recommend? Mine is Crucial - not fast nor expensive.

I wanted a full blown debian system - I found OpenWRT (etc) too limiting. It's not huge - 16GB would do. I went SSD because it should be more tolerant in a read/write environment.

Re TRIM - not sure about Linux. I'll have to look that up. Is that related to the new "SCSI Release block"[1] feature?

[1] So that underlying SANs know that they can truely delete and reclaim a block in a thin volume setup

Us too. I can max out all 8 CPUs in a compile and the disk just sits there idling, where spinning rust was running flat out. I often see the performance chart scaled to the gigabyte/second range - which means I/Os consistently in excess of 100MB/second. The best bit is that it multitasks so much better than a disc disc as the seek time is zero.

I understand that the trim helps, but is not essential. The disc here, for example, is 160GB. No way that's real, it has to be 192 with the 32 be>

For my home machine I bought Intel. Good reliability reports. Yes, I have a backup, but (a) not every day and (b) it's a PITA to rebuild.

Andy

I can do that on a humble twincore + SATA. the key is having enough RAM to cache the disk sectors..

We obviously have different compilers. This is MSVC2010, what are you using?

Andy

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required