Re: [abc] PLDI paper -99 cflow and steamloom

One thing to add to the discussion (Laurie's review nicely covers the
issues, I think):

In cflow, there are two kinds of overhead, at the update shadows,
and at the query shadows (our terminology). The steamloom technique
reduces the latter, but, it seems, increases the former. The other
techniques (i.e. the abc ones) tend to reduce both overheads, to varying
degrees depending on the individual technique. To its credit, the paper
discusses this issue somewhat. There's an interesting question which it
does not answer: in real world AspectJ programs (tm), is one form of
overhead more important than the other? This question is particularly
interesting for steamloom, since it favours one kind of overhead over
the other. A microbenchmark does nothing to answer the question. From
our AspectJ benchmarks, it seemed that both overheads were significant
in different benchmarks. Of course, our benchmarks are only a limited
sample of real world AspectJ programs, but at least they are more than
just a synthetic microbenchmark.

Ondrej

On Sun, Jan 08, 2006 at 07:30:41PM -0000, Oege de Moor wrote:
> It seems a weak paper overall.
>
> I'm somewhat suspicious of the choice of benchmarks. Why haven't
> they used at least some of ours? They can be downloaded...
>
> Indeed if it were accepted, it's essential that they distinguish
> between different kinds of optimisations performed by abc.
>
> Of course I'd also love to see more AspectJ work at PLDI that
> makes positive reference to abc, but I'd be surprised if this
> was judged up to the usual PLDI standards.
>
> -O
>
>
> > -----Original Message-----
> > From: Majordomo list server [mailto:majordomo@comlab.ox.ac.uk] On Behalf
> > Of Professor Laurie Hendren
> > Sent: 07 January 2006 15:43
> > To: abc@comlab.ox.ac.uk
> > Subject: [abc] PLDI paper -99 cflow and steamloom
> >
> > I have been reviewing paper 99 for PLDI, which is on the topic
> > of how the steamloom guys implemented cflow in their VM. This
> > is the only "aspectjy" paper that seems to have any chance.
> > However, there are certain problems with the paper. At least
> > I would like to at least clarify the discussion of abc in this
> > paper, even if it gets rejected - assuming that this paper will
> > find a home somewhere.
> >
> > I have prepared the attached discussion of abc, and used the
> > microbenchmark given in the paper to clarify the issue. I have
> > attached what I plan to add to my review (abcnotes.txt).
> >
> > I would really like to have abc'ers opinion on this paper. Do
> > we want to push it a bit - at least it shows that cflow is
> > a problem and abc does well ... or is it too weak a paper
> > overall? Do you agree with my discussion of abc/ajc and how
> > it relates to this paper?
> >
> > Cheers, Laurie
> >
> > +-----------------------------------------------------------------
> > | Laurie Hendren --- laurie.hendren@mcgill.ca
> > | Associate Dean (Academic), Faculty of Science,
> > | Dawson Hall, McGill University, 853 Sherbrooke St W,
> > | Montreal QC H3A 2T6 Canada, 514-398-7179, fax 514-398-1774
> > +----------------------------------------------------------------
> > | For contact and home page info as Professor, Computer Science:
> > | http://www.sable.mcgill.ca/~hendren --- hendren@cs.mcgill.ca
> > | Research: http://www.sable.mcgill.ca http://aspectbench.org
> > +----------------------------------------------------------------
>
>
>
Received on Mon Jan 9 14:43:44 2006