[linux-elitists] A modest proposal: SSD disk acceleration in kernel.

[Jim Thompson]

Eugen Leitl wrote:
> On Fri, Oct 24, 2008 at 12:37:41PM -0700, Greg KH wrote:
>   
>>> number of erase cycles that a flash can do?
>>>       
>> Physics.
>>     
>
> Sure, but even MLC does 10^5 cycles nowadays. Materials and
> processes progress continuously. 
>
> Incidentally, I'm doing an experiment tomorrow, by moving my 
> pfSense firewall on a WRAP from embedded to full install on
> a cheap Transcend 4 GB CF MLC flash. I'm hearing I can expect
> at least several years of half life at that usage pattern.
Do let me know how this goes, at least privately.  I could be convinced 
to move <what we sell that looks a lot like this> to
MLC CF cards.
>> Also note that the new Intel flash isn't the "traditional" type flash
>> that people have been using for years.  It's quite a different thing
>> down at the physical level with regards to composition of materials and
>> usage models.
>>     
>
> I've been hoping to see fully static (MRAM-based) designs by now. This
> doesn't seem to pull a bubble memory, but it's starting to smell a bit. 
>  
>   
>> There's going to be some pretty radical changes in the next 5 years with
>> regards to how memory and "flash" start getting intermixed.  Don't worry
>>     
>
> Will we get GC on persistent objects, and no more filesystems? And
> just how well is Linux is going to play with that, I wonder.
>   
See, its all gonna cycle back around to a lispm.   Perhaps with 
python/perl/tcl/php re-written as DSLs in lisp.
You'll get your open/close/read/write/seek semantics, but the 'files' as 
such ... won't be files, at least not
in a filesystem.

Linux might be at its core, but 80% of the 'userland' would evaporate.  
not too different from Android, actually.

Jim