> I have read that SSDs have a much longer lifespan, and are a lot faster to
> boot (pun attempt).

Completely false on both counts, depending on circumstances and subtracting
the pun.

SSD's are _usually_ faster, and don't have mechanical wear mechanisms.  But
there are also two basic facts of physics working against you:

1) The fundamental electron tunneling mechanisms used have a
wear-equivalent behavior, and they literally slow down with every erasure.
Depending on the technology, at a 10:1 ratio or worse.  When near EOL, some
devices can take 1/2 second or more PER ERASURE to finish.

2) Some flash technologies have a wear-out threshold in the hundreds
(10^2!) of cycles.  The best ones are only millions (10^6).  10^4-10^5 are
the middle of the road.  Most flash vendors don't want to tell you the raw
numbers.  The newest high-density MLC/TLC flashes are the worst.

Plus, there are more factors:

3) Wear-leveling mechanisms, required in order to compensate for the above
two conditions, are often not matched to the filesystem.  This can result
in highly-unoptimal use behavior of the raw flash device, causing
accelerated wear on the flash.  Yes, making things worse instead of
better.  Current trends are to integrate this necessary mechanism into the
device, where it is opaque, unspecified, and unadvertised.  By the time you
find out it's actually shit, it's too late.  Low-cost commodity devices are
usually the worst.

4) The more full the SSD is, the less free space there is to push around
for wear leveling purposes.  A nearly full device will wear out much faster
than a nearly empty one.

5) Retention is usually unspecified, but can be measured in months on the
worst devices.  Higher temperatures exacerbate the problem.

For some heavy write-oriented applications, a traditional rotating disk
will actually work better and last longer!  A company I once worked for
changed production from rotating to SSD, "to enhance reliability."  Failure
rates in the field went _up_!  They did not switch to a flash-friendly
filesystem format, and the traditional fs just wore the flash out in short
order.  The 'improvement' was made by the purchasing/marketing department,
and engineering was neither involved nor consulted.  Oops.

OTOH, a truly optimal implementation, largely read-only and well-matched to
the application, will really fly, last a long time, and will be immune to
physical abuse.

Our big Macs here at home use SSD for the boot/application disk, and
rotating media for all the user files.  They're blazingly fast in general
operation, but a catastrophic SSD failure will not result in the loss of
critical data.  These machines all use Time Machine to back up to a server,
which uses 2TB RAID HD's to catch.  One of these raw HD's is rotated out to
off-site storage quarterly, and the RAID is rebuilt with the oldest offline
device.  (4 2T mechanical hard drives in rotation, 2 serving 'live' at a
time, 2 in separate off-line off-site locations.)

Fast, and safe.

-- Jim

To search list archives http://www.okiebenz.com/archive/

To Unsubscribe or change delivery options go to:

Reply via email to