Anon (no@email.com) on 1/25/11 wrote:
---------------------------
>Michael S (already5chosen@yahoo.com) on 1/25/11 wrote:
>---------------------------
><....
>>It seems to me that it also useless when
>>a) FS manages free logical blocks in LIFO manner (which seems to be the only logical option anyway).
>
>1 - this relies on the controllers pulling remap tricks to save you from HORRIBLE performance on write.

No, that is required anyway for NAND.

>2 - you are going to have fun keeping those LIFO tables persistent across powr outages, etc ;)

That does not require anything more special than a first-block allocator. In practice you want a bit more intelligence, even for SSDs (smaller extents and fragmentation don't hurt as much on SSDs, but still preferable to avoid it). But it does not have to be absolutely perfect LIFO and preserved over power fail etc.

>
>>b) Used space rarely shrunk by more few per cents of total volume and the average trend is up.
>
>You are looking for cases where a sensible feature does not have an advantage -
>its not hard to find cjases where memory caches also dont work, should we throw those away also?

They are not thrown away. There are options and defaults that can be changed, and there are ways to do it offline. The important point is that the defaults are right for the most common usage.

>
>in the real world, and for a good implementation, TRIM is a big advantage often,
>and no disadvantage ever (ignoring stupid implementations)

Let us see your numbers.