Operating Systems in Memory (I Wonder...)

[David Lloyd]

Richard et al,

>> If you want your disk access to go fast, buy more RAM to make the
>> cache bigger.  Don't do what most people do, which is to buy enough
>> RAM to hold your currently running applications.  Buy enough RAM to
>> hold your currently running applications AND their datasets.  Then
>> it'll be fast.  Greased lightning.  No additional work required.

I agree that the various kernel level caches that exist will eventually 
cache the data (executable and datasets).

> Got to agree with you, Mark. Most of my tuning experience has been on
> mainframes, where it's BEST to allow the paging system to look after
> things. Paging IO was so optimised, no other types of IO came even
> close.

This may be the case, however...

> And if you have enough memory, it'll never fill up, and eventually,
> everything you need will be there, after the first time.

Maybe another way of saying this is: I want to replace my  hard disks 
with something REALLY fast. Unless I'm missing something, hard disks 
such as:

  * www.seagate.com/products/consumer_electronics/ld25_series.html

...for read (and write) are still slower than DDR2 RAM. So:

  * Is the PAGE I/O / Buffer Cache going to run faster backed by:
    - hard disk storage; or
    - RAM storage
  * Will find / run faster:
    - using a hard disk; or
    - using RAM with tmpfs (or ReiserFS on a Ramdisk)

Whilst I agree that leaving the caching to the experts is probably a 
Really Good Idea (rtm), however I would suggest that it's not THAT 
difficult to conclude that accessing RAM is faster than accessing Hard Disk.

In fact, I suspect that the caching experts actually agree with me.

DSL

[i.e. WHY would they bother developing in-memory caches if they could 
access what was on a hard disk just as quickly?]