"Juan J. Quintela" wrote:
> Daniel Phillips writes:
>
> > Phase tree is my terminology. In my previous post I failed to mention a much
> > bigger reason why phase tree is not shadow paging: phase tree does not use a
> > shadow page table. Instead, it uses a tree where the differences between one
> > commit phase and the next generate a minimal set of new nodes, and only these
> > need to be written to disk.
>
> If you have a good implementation of shadow paging, you are supposed
> to write to disk the minimun amount of data possible, i.e. only the
> things that have changed/are new. The method of writing the new data
> to free blocks and then change only the root is a known and used
> mechanism in shadow page implementations.
Yes, but that's like saying the method of sorting by exchanging is well
known and you are supposed to do the minimum number of exchanges,
therefore heapsort is the same algorithm as quicksort.
There are indeed some important similarities between shadow paging and
phase tree: the update is atomic and changes are written into
freespace. There are also some important differences: phase tree
specifies that the entire filesystem is to be held in one tree
(including all allocation and freespace metadata) and specifies exactly
how the copy-on-write is to be done. It also provides a third metaroot
to avoid blocking while the filesystem is being updated. I don't doubt
there are other significant differences as well - I'll try to keep a
list as I work on my paper.
-- Daniel- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Aug 07 2000 - 21:00:09 EST