wrstuden@nas.nasa.gov (Bill Studenmund) writes:

>On Fri, 21 May 1999, Manuel Bouyer wrote:

>> On Thu, May 20, 1999 at 09:23:45PM +0400, Oleg Polyanski wrote:
>> > 	they are very different. xfs is an ordinary (but extremely optimized
>> > for high speed data access) journaling file system and stores the transaction 
>> > log somewhere on the disk (in other slice of disk or in the slice of another
>> > disk or inside the filesystem itself). lfs is a log-structured file system,
>> > there is no transaction log or data - file system itself is a log. lfs is
>> > good in many terms but unfortunately it is write only file system; data
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> > reading without notable performance impact is impossible. lfs is an example
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> > of excellent research work but in real life it is unusable.
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>> 
>> Hum, the few tests I ran on LFS showed it to be faster than ffs for reading
>> as well. Do you have some numbers ?

>Someone else in another discussion of LFS described the point I think Oleg
>was refereing to.

In the text you qouted he stated in no uncertain terms that LFS is
unable to yield decent read performance and that it is unusable in real
life.  Obviously, those claims were made without any practical
experience.

>That a highly fragmented file will read very slowly on
>an LFS (as the data are in multiple segments). That doesn't mean reading
>is slow, just a highly-fragmented file will read slow.

Reading a highly fragmented file from *any* filesystem is slow.
That proves nothing.

As for your hidden claim that LFS produces a significantly higher then
average percentage of highly fragmented files than FFS.  I don't notice
that in _my_ usage.  Of course, no one has reproducible numbers on that
yet because virtually no one could get those numbers until Konrad fixed
the code.

-- 
Christoph Badura					www.netbsd.org

	Anything that can be done in O(N) can be done in O(N^2).
	-- Ralf Schuettau (after looking at a particular piece of code)