[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[RFT] Fixing performance problems.



I got rid of the linked lists of fragments, so append to and read from a
large file should now be an O(log n) operation not O(n). We also started to
use the 'point' interface to avoid pointless memcpy() from flash. That needs
more thought before it's really safe w.r.t simultaneous writes/erases though.

And I finally ditched the CRC checks on both date nodes and their payload
during mount, instead doing them at runtime (or in the GC thread) before
allowing GC to happen or at the time the inode to which they belong is
required.

Guinea pigs welcome. Please get the latest code from CVS and beat it up.

I've now started to mark physical nodes as 'pristine' when the GC can copy 
them intact, so we can avoid the decompression and recompression which 
currently happens, and even the read_inode(). That should dramatically 
improve runtime GC performance too, but the mount time was annoying me most 
so I did that first.

--
dwmw2



To unsubscribe from this list: send the line "unsubscribe jffs-dev" in
the body of a message to majordomo@xxxxxxx.com