> On Aug 28, 2020, at 1:40 PM, Al Kossow via cctalk <cctalk@classiccmp.org>
> wrote:
>
> On 8/28/20 10:10 AM, Paul Koning via cctalk wrote:
>
>> SD is a packet based storage device on a serial interconnect
> You really do need SMART monitoring on solid-state storage
> which may or may not exist in the adapters. SSDs will silently
> fail if they run out of sectors to write to.
Yes, if they wear out. But the lowest I have seen is around 100 writes per
sector, which means that once you've written 100x the device capacity total
(times the "write amplification" which depends on write size patterns) you'd
start to consume spare sectors, and you don't have a problem until those run
out which is still some time later.
That said, I have seen it happen, with very small CF cards and software that
was, by a coding slip-up, writing every few minutes non-stop.
> Also, I discovered recently that there is a maximum number of hours
> measured in years on SSDs and systems will start throwing SMART
> errors when that is exceeded. I have a few doing that now on systems
> with minimal writes but lots of hours.
That's curious. There may be a read retention time limit, though I think that
applies only if the device is powered down. When powered up, the device
firmware takes care of refreshing the sectors, somewhat like memory refresh but
with refresh cycles of weeks or months.
> There are long discussions elsewhere of the dangers of using non-industrial
> rated CFs and SDs in storage applications.
Good advice for serious work especially if you're running heavy workloads. I
wouldn't worry about it much for my home firewall. But the servers running
bitsavers.org are an entirely different matter.
One way to look at it: if you are, or should be, using RAID, you should be
using industrial grade SSDs.
paul
