[ale] SWAP over NFS

Brian Pitts brian at polibyte.com
Thu Apr 17 16:54:01 EDT 2008


JK wrote:
> Brian Pitts wrote:
> 
> I didn't know the cards did wear-leveling on their own; that's
> a nice feature (and makes me wonder if some schemes I've seen
> implementing explicit wear-leveling on the OS side are gonna
> end up shooting themselves in the foot by interacting badly
> with the card's on-board wear-leveling mechanism).

The various log-structured filesystems are for flash devices that 
provide a raw flash interface. All the consumer stuff should do block 
mapping and present an ATA interface.

  But in any
> case, when wear limits are hit, one would start seeing random
> mysterious errors due to corruption of the swap device, with no
> apparent cause.  This seems to me a Very Bad Thing. 'Course,
> same thing can happen with bad HD sectors, but my impression is
> that this situation could occur much more rapidly with flash,
> especially if (as in the scenario in this thread) the whole reason
> for adding the flash device is that you expect the system to
> swap its ass off all the time.  I mean, 1,000,000 is really
> a pretty small number.

Let's assume I can write 10,000,000 bytes a second to a 1,000,000,000 
byte CompactFlash card that can survive 1,000,000 writes. (capacity * 
writelimit) / writespeed equals 100,000,000 seconds, or over 3 years, of 
continuous writing assuming perfect wear leveling. I think the user will 
outgrow webkinz by then.

-Brian


More information about the Ale mailing list