Aaron Spink (aaronspink@notearthlink.net) on 1/19/11 wrote:
---------------------------
>anon (anon@anon.com) on 1/19/11 wrote:
>---------------------------
>>Which mapping tables are these, and where are they stored?
>>
>They are the mapping tables between the logical interface presented to the storage
>controller(LBA, etc) and the physical block/page/device location within the SSD.
>All modern functioning SSDs have them. Total size and granularity of the tables
>are primarily a function of interface block size, SSD size (aka 80GB/160GB), power
>loss to power fault time (aka someone pulled the plug on the server/hit the power switch), etc.
>
>The various schemes used to store the tables vary greatly with controller design.
>Some controllers use small on chip sliding tables and back fill to reserved flash
>storage space. Some use high granularity on chip and query tags in flash for intra-block/page
>information. Others use external SRAM/DRAM/PRAM to enable larger table, faster access, and better granularity.

SRAM/DRAM? Battery backed? I'd be surprised if that was the case, on a commodity SSD.

Anyway, right. So when you say "it's free, it just has to update this table", you mean that an IO has to go down the wire, and the device has to do some processing and potentially some banging on NAND (or PCM or whatever) or even this SRAM, and then send the completion back up.

That's only "free" for some restricted definitions of "free".

>The mapping table scheme and GC schemes largely are the differentials in SSD(flash
>based) performance given that almost everyone is using the same number of channels
>and same speed flash ICs. The highest performing devices (aka enterprise random
>access small block size which is the hardest case) tend to rely on external DRAM,
>512B granularity, and capacitors. It is also possible to go external DRAM, 512B
>granularity and journal/rebuild on power on but much more complicated algorithm wise.

And even on those, TRIM will be a total loss if the filesystem rewrites the same LBA before the old one is garbage collected.

So when you say "pretty obvious", this is also only true for some definitions of obvious.

And that's why doing trim in the filesystem is hard.