This note mechanically notifies you of the status of your submission to
the Winter, 1995 USENIX Conference.  It was generated by a program.

We have reviewed your abstract (our number 041) with the title:

File System Logging versus Clustering: A Performance Evaluation

and are pleased to inform you that it has been accepted.  An author
instruction packet will follow later on.

We received 80 submissions and were able to accept only 27 of them.
Congratulations on writing one of the ones we liked!

If for any reason it will not be possible to present your paper at
the conference, please notify me at once.

Committee member Noemi Paciorek will be contacting you about your
paper to help "shepherd" it through the final publication process.
You are expected to provide your shepherd with a near-final draft of
your paper by October 1, 1994 (or a later date chosen in consulation
with your shepherd) in order to provide time for additional reading and
comments.  Your shepherd has been assigned to ensure that you meet some
of our concerns raised during the reviewing process, and we ask that
you cooperate.

Below are (anonymous) referee's comments about your paper.

One reviewer noted:

    1: Is the work of interest to a reasonable portion of the USENIX
       attendees?  If not, please try and suggest an alternate forum
       for the work.

       Yes; real data on LFS performance are valuable.

    5: What are the weakest and/or least understandable parts of the
       work with respect to its presentation?

	Figure 4 is hard to read. Better choce of linestyles
	would help.

	Also, page 9 seems to introduce the term EFS; I searched
	for it on prior pages without success.


Another reviewer noted:

    1: Is the work of interest to a reasonable portion of the USENIX
       attendees?  If not, please try and suggest an alternate forum
       for the work.

	This work further compares FFS and LFS and should be of
	interest to those following this file system controversy
	as well as those interested in these two fundamentally
	different approaches for organizing data on disk.

    3: Are the references sufficient?

	Yes, but the canonical FFS reference should be
	included.

    4: Is the English use and presentation of this work acceptable?

	Some critical text defining the cluster layout and what
	goes in the gaps of a FFS is lost at the bottom of page
	4.

    5: What are the weakest and/or least understandable parts of the
       work with respect to its presentation?

	In Figure 4 why no 90% figure for LFS without cleaning?

	In Section 4.2 I'm not sure small file I/O can be
	dismissed so easily.  While most I/O may involve large
	files, I'll argue that it is poor performance on small
	files that is most keenly and repetitively felt by the
	user.

	In general the figure captions and running text differ
	slightly on the interpretations of the results.  For
	example, the caption of Figure 5 says performance
	improves 50-75% while the running text uses 50%.

	In Figure 5, it would help to sort the legend on the
	right in the same top-to-bottom order as the graph
	curves.

    6: Is the work technically correct?

	In general the investigations of the causes behind the
	various measurements are very thorough.


Another reviewer noted:


    Please use a serif font for the final paper; it will be much
    more readable.  Be careful to match the person of your verbs
    with the word "data" which is a plural noun.

    sec 1.0, bullets:  The first bullet's wording was
    confusing.  How can performance impact sequential access?

    sec 2.0: How many directory blocks did you end up with per
    directory with your 100-files-per scheme?  Are you running
    4.4BSD-Lite straight up on the sparc?  I thought it was
    incomplete (or are you using something like NetBSD?)

    sec 4.3:  Do you find any difference between fragmentation
    rate if the write rate to a filesystem is slower?  I suspect
    that /usr is written to far less than /home, for example.
    If you have anything to say on this issue, please do say
    something!

    Figure 5:  It'd be easier to read this figure if both glan-6
    lines had the same shading, and both glan-8 lines had the
    same shading; then you can use circles or boxes to indicate
    bw or score.  section 5:  If you could use LFS for a USENET
    news partition, and report back, that'd be great!


Another reviewer noted:

    Nice paper.

       The abstract clearly states the goals and conclusions,
       albeit in a somewhat lengthy fashion.

       The work was clearly described.



Thank you again for your submission.  Please contact me if you have
any questions.

I hope to see you in New Orleans!

					Peter Honeyman, Program Chair
					+1 313 763 4413