[Beowulf] SSD caching for parallel filesystems

[Ellis H. Wilson III]

On 02/09/13 16:32, Bill Broadley wrote:
> On 02/09/2013 01:22 PM, Vincent Diepeveen wrote:
>> SATA is very bad protocol for SSD's.
>>
>> SSD's allows perfectly parallel stores and writes, SATA doesn't.
>> So SATA really limits the SSD's true performance.
>
> SSDs and controllers often support NCQ which allows multiple outstanding
> requests.  Not sure if that fits the definition of "perfectly parallel",
> but it does allow for outstanding requests to be answered in the order
> of the devices choosing.

Very true, and more importantly, there are multiple internal SATA 
controllers, making this a parallel setup once the requests make their 
way "into" the SSD.  I agree on the premise that SATA is less than ideal 
for NVM memory in general, but I'm afraid I won't go so far as to say 
SSDs are "perfectly parallel" devices.  These things have many dies 
within a package, but with limited buses to get at each die.  Similarly, 
their are many packages along each channel, but with limited channels to 
get at each package.  This all boils down to contention on the 
channels/buses for requests and brings to light just how important it is 
to choose things that aren't busy at this time.

Perfectly parallel would be if every die (really, every wordline) had 
it's own bus.  Now, that would be fast :D, and horribly expensive.

Best,

ellis