Re: [boost] [serialization] performance comparison

From: brass goowy (brass_at_[hidden])
Date: 2007-03-01 23:43:11

Martin Slater writes:
>> 
>> We haven't yet been able to provide support for the build integration

>> that you mention, but we've thought about it and hope to deliver 
>> something there. 
>
>Anything that involves the build process contacting a webserver is 
>unacceptable as there is just far to much that can go wrong inbetween 
>that would mean a build cannot be done. 

I agree it is a tenuous system.  On the other hand how much of a leap
is it from using a network file system when building?  There is usually 
more distance involved, but in both cases requests and replies are 
made.  People around the world work on maintaining and improving 
the internet and it isn't like reliability is needed just for online code 
generation.

>And I've yet to see any indication you support even a fraction of what

>boost serialisation does, versioning and a fully general system for
user 
>defined types and archives the first things that come to mind, 

In the past, differences between the B.Ser and EE approaches to
version support have been discussed in this group.  
http://lists.boost.org/Archives/boost/2006/05/104748.php
http://lists.boost.org/Archives/boost/2006/05/104882.php

I don't think there is anything about the EE approach that 
prevents doing what B.Ser does with archives.


>speed is 
>only a part of the equation and not something I have run up against
with 
>boost serialisation that a small amount of work on a more custom
archive 
>has solved (which was similar to the work done that is now in 1.34). 

No one has answered my question of how to obtain 1.34.  I tested with
1.35.  It was better than 1.33.1, but still more than three times slower 
than the EE approach in the list<int> test.

>That's not to say I'm not interested in helping to find places where 
>boost.serialisation could be faster, hence why I'm in this discussion.

What Ramey said earlier in the thread about possibly ditching
streambufs makes sense to me.  Beyond that it would be helpful 
to be able to review the code that compilers generate in a 
simple B.Ser use.  I don't know of any compilers that have an 
option to reveal that though.  You could compare the assembly code, 
but it shouldn't have to be done at that level.


Sholom,
Brian Wood
Ebenezer Enterprises
www
_______________________________________
your personal webtop. @ http://www.goowy.com