Woodhouse on flash storage

[John Watlington]

Coincident with the industry-wide change to managed Flash is the change
to smaller transistors and storing multiple data bits per cell (MLC,  
vs. SLC or
single bit per cell).  This has brought a huge increase in error  
rates, and a
decrease in reliability and erase cycle lifetime.

The bad news is that even SLC Flash reliability has suffered with  
decreased
transistor sizes.  For example, the SLC NAND Flash used in XO-1 is  
rated for
100K erase cycles per block, while newer SLC parts are rated for 10K  
cycles.
A typical MLC part is only rated for 5K cycles.

The good news is that the new wave of controllers designed to manage MLC
devices seem to be relatively good, and pricing is much better.   At  
this
point, an MLC device is cheaper than an SLC device with half the  
capacity.

As a number of people have pointed out, there is little incentive for  
the
makers of the FTL chips to use Open Source.   The efficiency/speed/ 
reliability
of the controller firmware is the differentiator between them ---  
supporting
FOSS commoditizes their product even more.

UBIFS has performed v. well in testing.   Unfortunately even Nokia, the
company that has been funding its development, seems to be moving
away from it in their newer products.

Cheers,
wad