JFFS3 & deletion dirents RFC

Artem B. Bityuckiy dedekind at infradead.org
Fri Mar 25 03:36:39 EST 2005


On Thu, 2005-03-24 at 14:01 -0800, Jared Hulbert wrote:
> > We can't get rid of version in data nodes.
> 
> I'm not suggesting we get rid of 'version' in the actual nodes.
> 
> As I understand this adding 'version' to struct jffs2_raw_node_ref
> increases the RAM usage by N*(X+Y) where:
>   N is the size of the version field
>   X is the number of direnties
>   Y is the number of data nodes
> 
> It seems there "ought" to be a way to only increase RAM size by N*X. 
> I'm assuming Y>>X so this could be a big factor for RAM usage.
> 
> Am I missing something?
> 
>  ,Jared
Ok, now I understand you. Yes, good point, thanks. Frankly, I bear in
mind 2 more things which I didn't publish:

1. We will have some memory "reap" or "shrink" mechanism in JFFS3
(possibly with the ICP help), this is why I so bravely increase the
node_ref's size.
2. I strongly suspect that the version field well be very useful for
other purposes as well, even in the inode node (AKA data node)
node_refs. This is why I don't afraid to add version to node_refs of all
node types.

So, I still need to design more precisely that "shrink" mechanism and to
prove other usage of (node_ref)->version.





More information about the linux-mtd mailing list