About SSD's, flash, NAND, writing

[jimhsu]

I had some technical questions about flash memory that the OCZ forum was not really able to answer: here

Here goes...

As it is defined, flash memory is "0" in a programmed state and "1" in an erased state.

http://img23.imageshack.us/my.php?image=flashnwz.png

This is true for both NAND and NOR flash. SLC only has these two states, while MLC has four:

Value State
00 Fully Programmed
01 Partially Programmed
10 Partially Erased
11 Fully Erased

(reference)

Now, flash memory has to be programmed to write bits. To do so, the bit must be in an erased state. A block can be read or written (programmed) on a page basis, but erasing must be done by a block basis (see the anandtech article). Hence, drives achieve a "steady state" performance upon which any additional written data has to trigger an erase command first (a bit simplistic, but it'll suffice for now).

Questions:

1. How does the SSD know that a page is writable (i.e. fully erased)? Intel says:

"SSDs all have what is known as an ?Indirection System? ? aka an LBA allocation table (similar to an OS file allocation table)...."

If this is true, then writing any erase pattern (zeros, ones, or a mix) with something like Eraser will not restore "free space" to programmable space, because each block is "addressed" as erasable or programmable by the indirection system. Now if there's some way to access this indirection data, then this should not be a problem. If I am reading this wrong, then writing 1's in eraser should restore the drive to pre-steady state performance in free space regions.l

2. Related: How does TRIM differ from simply writing a bit pattern to a particular block in the SSD?

3. Isn't it horrendously bad for this indirection region to be located on flash itself? Whenever a block is erased or programmed, that block has to be written, AS WELL AS the information in the indirection region that indicates whether or not that block has been erased or programmed.

Can anyone try different bit patterns in Eraser (or similar) on a "steady state" SSD to confirm some of the above observations?

 

[jimhsu]

OK This is more related to the Vertex, but something between anandtech and PCPer is directly contradicting:

Crossposted from OCZ forums:

Ok, but doesn't the Vertex exhibit significantly worse degradation when in a steady state condition? At least that is what the anand article suggests for the delta between "new drive" random write bandwidth and steady state bandwidth. Then again they're using an older firmware revision.

I know that all flash SSDs need to erase blocks before writing data on a "full" drive, but the article suggests there is something else at work with the Intel drive that Intel categorically denies (http://pchardwareblips.dailyradar.co...d_performance/). Notice that anand was also unable to replicate PCPer's performance problems not true, see below. Additionally anand claims that steady state does not affect random write performance (http://anandtech.com/storage/showdoc.aspx?i=3531&p=14) suggesting that PCPer's observed effect is "something else".

PS From page 15: "There?s not much I can say about the issue other than I?ve been working with Intel on it very closely and it?s not something I?m overly concerned about at this point. I can replicate the PC Perspective data but not by using my machine normally. Through the right combination of benchmarks I can effectively put the drive in a poor performance state that it won?t recover from without a secure erase. I should also mention that I can do the same to other drives as well."

Here the article directly contradicts PCPer's observation that this special form of degradation is limited to the Intel:

We did not mention other SSDs because the entire premise of the article is that the Intel SSD shows the slow down while every other SSD benchmarked by PCPer thus far does not. This includes several of my own personal drives (Memoright, Samsung, etc).

Investigative journalism is tiring...