Re: [Boost-users] [Boost.MPI] question about serializingcustomobject

Subject: Re: [Boost-users] [Boost.MPI] question about serializingcustomobject
From: David Abrahams (dave_at_[hidden])
Date: 2010-08-27 14:13:16

At Fri, 27 Aug 2010 09:44:39 -0800,
Robert Ramey wrote:
> 
> David Abrahams wrote:
> > At Thu, 26 Aug 2010 16:05:41 -0800,
> > Robert Ramey wrote:
> >>
> >> Dave Abrahams wrote:
> >>> BoostPro Computing, http://boostpro.com
> >>> Sent from coveted but awkward mobile device
> >>>
> >>>> If one were using heterogenious machines, I could understand the
> >>>> usage of MPI types. But as I understand it, the MPI serialization
> >>>> presumes that the machines are binary compatible.
> >>>
> >>> You're mistaken.
> >>>
> >>>>   So I'm just not seeing this.
> >>>
> >>> Start reading here:
> >>> http://www.boost.org/doc/libs/1_39_0/doc/html/mpi/tutorial.html#mpi.skeleton_and_content
> >>> and all will be revealed
> >>
> >> lol - I've read that several times.
> >
> > I always wonder, when you write that, whether you're physically
> > laughing out loud.  That's OK; don't spoil the mystique ;-)
> 
> I actually do think I snicker a little bit.  Sort of one "snick".
> 
> This above is exactly the type of thing which provokes this reaction.
> Reading your comment, one has to conclude that
> 
> a) you think I haven't read it but am expressing an opinion anyway
> b) that you think that the documentation is clear and complete
> c) that I'm somehow responsable for not finding/reading/understanding it.
> 
> You make this kind of presumption all the time.  

It's a weakness of mine, I admit, for which I apologize.  But I wasn't
doing that in this case.  I was a bit terse, because as I noted above
the message was sent from an awkward mobile device.  Actually,

a) I did think you hadn't read it, but I was trying to be helpful by
   pointing at the part of the doc that relates to your question.  I
   thought you were saying "I don't get it; please help me
   understand," not that you were "expressing an opinion."

b) I haven't read all the documentation.  I did presume it was
   complete, but not that it was clear, hence
   http://groups.google.com/group/boost-list/msg/d14a38f3c55613b6

c) I don't see how anything I wrote could lead to that conclusion.

> It doesn't bother me but it always provokes at least one "snick"
> 
> >> I just never found it to be very revealing.  The word skeleton
> >> seemed pretty suggestive. It's still not clear to me how such a
> >> think can work between heterogeneous machines.  For example, if I
> >> have an array of 2 byte integers and they each need to get
> >> transformed one by one into a 4 byte integer because that's closest
> >> MPI data type,
> 
> > I think you don't understand what MPI datatypes do.
> 
> This is true.  I suppose that's one reason why the documentation
> made no sense to me. They just looked like special types
> to make identify primitives accross differing architectures.
> 
> <snip>
> Interesting information which one should consider adding to the
> MPI documentation.
> </snip>

Actually, I wish you had given me some detailed feedback.  I was
thinking of writing it up in such a way that the Boost.MPI docs could
point at it, but I can't tell whether it connected for you or not.
> 
> > And since Boost.MPI and Boost.Serialization are so closely
> > related, I think it's especially important that *you* underestand.
> 
> I disagree.  Boost MPI depends upon Boost.Serialization
> but not the other way around.  

I didn't say they were interdependent, just that they were
closely-related.  But anyway, I think you're being shortsighted: The
success of tools built on top of Boost.Serialization is an important
indicator of the correctness and genericity of its design.

Boost.Serialization *is* dependent on Boost.MPI for a portion of its
userbase, and to the extent that you are interested in supporting that
portion of the userbase and enabling anything like Boost.MPI to exist,
the things can/should do to Boost.Serialization depend on the
requirements of Boost.MPI.  

> I shouldn't have to understand Boost MPI just as I can't be expected
> to understand all the usages to which boost serialization might be
> applied.  

Welcome to generic library design! :-) You have to decide what your
application domain(s) is/are.  Do you want to serve people who are
trying to save/load their desktop applications' files?  Do you want to
serve people who want to save/load XML?  Do you want to serve people
who want to checkpoint the state of long-running calculations?  Do you
want to serve people who are trying to implement RPC?  etc...

> Trying to do this, aside from the time involved, might
> well be counter productive in that it can trick one into coupling
> things which would otherwise not be.  For all these reasons I've
> refrained from investing a lot of time in understanding MPI as it
> relates to serialization.  I'm happy with Mathias efforts and
> commitment to supporting his library and don't want to muck up the
> works.

You can't be a generic serialization library without reference to real
applications.  Maybe you still haven't really decided whether serving
Boost.MPI and its users is something you want to do.  I think that
might account for a good deal of the recurring friction we experience.
It would be a good idea to settle on an answer to that question, along
with the question of what other applications you're willing to
support.

> I only really have few observations/suggestions at this point.
> 
> a) It would be  helpful if there were a way to test the serialization
> of the archive classes in boost MPI without having MPI installed.
> If this is not possible, it would seem to me that the serialization is
> intertwined with the data transport - rather than be separated
> as it is in the stream/streambuf io/stream design.  This would look
> like a design flaw to me.

Helpful to whom?  I agree this is all a good idea, but I'm not sure
how this relates to anything else we're discussing.

> b) user experience seems to show that archive
> construction/destruction is a significant performance issue when a
> new archive is made for each data transmission.  On the other hand,
> one has to do this since the current archive implemenation track
> addresses of serialized obects so the same archive can't be use send
> the same structure (maybe with changed data) multiple times.  

I can't understand how you reach that conclusion.  In my long
explanation I thought I made it clear that Boost.MPI and its database
get a great deal of its performance advantage from exactly that:
sending the same structure multiple times with the same MPI type map.

> Given that MPI has a focus on performance, I wonder if this has been
> considered. I looked a the documentation, code and examples and it
> wasn't obvious to me how this is question was addressed - if at all.

We've obviously misunderstood one another somewhere along the way.  It
would be good if we could get that cleared up.

> c) I think the above information regarding how MPI and serialization
> fit together in boost MPI would be a worth addition go the MPI
> documentation.  AND it's already written !

It has been my intention to get it out in a more public place than
this mailing list.  Again, detailed feedback would be helpful.
It can't already be perfect, or the misunderstanding cited in b)
wouldn't have arisen.

> You should know that all your efforts to educate me are not wasted.

Glad to hear it.

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com