EXT4 on SSD - Page 2

Chilly Willy · #16 14th November 2010, 06:19 PM

Yeah, I know that MTBF is just a lab guess & I don't totally buy into it either as there is only ONE WAY to actually KNOW FOR SURE & THAT is with REAL WORLD data, but the device hasn't been around long enough & in a few years who know what we'll have. I was just pointing out the fact that THEY make such a claim & from it, it SEEMS that these things SHOULD last a few years or the life of the machine. Again *IF* there data is true. (there claim, NOT MINE!)

As with ANY device, it can last a long time or it can fail immediately so this can be argued until the cows come home but it's STILL a crap shot either way. - personally, I MAY take a shot at SSD's & see... (call it luck if you like, but I USUALLY get MORE usage than I should out of electronics anyway - & I'm not doing anything to cause that, either)

jpollard · #17 14th November 2010, 07:50 PM

You weren't paying attention.

To generate a MTBF they use REAL WORLD data.

They write to the device as fast as it will go.

They write until it fails.

The advantage the lab has is that they can also plan other stress on
the device.

Over heat, freeze, overpower, under power... power surges/spikes.

I usually get more out of my disks too - but I also know I'm not stressing
them as much as the MTBF assumptions.

I've got memory sticks that are 5 years old that still work just fine.
But I also know I'm not using them as a root filesystem.

Flash generally isn't designed for that. That is why things like "noatime"
and disabling journaling are so important. They help minimize the
writes done to the filesystem, which in turn, minimized the chance of
failure on flash.

It is more unlinkely that a flash device fail immediately than a disk - it
is less sensitive to vibration, storage, and shipping issues. If it passes
the read/write tests at the end of manufacturing (which primarily tests
the onboard controller, and not the flash) then it will work when delivered.

But ignoring the failure modes will get you in the end.

Chilly Willy · #18 14th November 2010, 08:18 PM

It's not that I'm not paying attention, it is a difference in usage of the term. By REAL WORLD I mean as they would be used by people over a given time, in ALL sorts of manner & that includes being "abused" - not some sort of lab "stress test". - look at the data for ANY electronic device & we would STILL be saying the same things. - All this data is just a reference - when it comes to devices as such, there are NO absolutes.

BTW: why am I getting "called out" on this? I only made an observation, I'm not the one that called it a "MYTH"!
I don't want to argue, & I don't want to debate. So as far as I'M concerned, enough said...

stevea · #19 15th November 2010, 04:15 PM

Quote:

Originally Posted by jpollard

Nope - MTBF is valid. But you need to realize that to get that
figure they first run samples of the chips (random selections)
and run them to failure. In the lab that may take a few weeks
(or just a few days). Now they have a total number of writes
per time (whatever it was) until failure.

Not usually. Most MTTFs are created by a combination of test and calculation based on individual component failure rates. There is no ability to do the sort of multi-dimensional testing that you suggest (vibration, heat cycling, power cycling...) and it would take vast resources to do so. Instead they do a number of single dimensional tests and calculate the stats for the ensemble. In the case of SSDs it only takes about 24hrs to hit the write-limitation wear limit on a 160BG drive, but that can easily calculate the limit based on single chip write limits and the wear-leveling algorithm. OTOH to do 50,000 power cycles might take 3 days to a week, and that's just the mean figure, some units may take many months to fail, so they likely create an error bound in the power-cycle estimates.

Most drives have no temperature spec aside from a recommended and warrantee limit range, yet it's probably a major factor in mechanical disk failure. Mechanical disk drive MTTFs are notoriously inaccurate (Google has published several papers on the topic) but they aren't really simulating the power supply variations, temps, airflow, air pressure, and mechanical shocks of the real-world. It's not that the MTTF is a lie, it's that it only applies to the Mfgr's lab conditions.