Re: [Boost-users] Lite serialization lib optimized for performance

Subject: Re: [Boost-users] Lite serialization lib optimized for performance
From: Robert Ramey (ramey_at_[hidden])
Date: 2008-12-15 12:39:17

"I think boost::serialization will find it hard
to beat with its current design. Except for a smaller header which
currently only keeps track of null pointers archive size is pretty
much equal to that of a hand crafted solution."

Why would that be?  as far as I know an archive contains
absolutely no data which can be eliminated and still have the
system work.  Where is the redundacy in the current archives?

Robert Ramey

Sebastian.Karlsson_at_[hidden] wrote:
> I agree, I however tried to first achieve the desired results by just
> extending boost::serialization, but without much luck.
> boost::serialization is a very large library in terms of source code
> and it's pretty hard to wrap your head around the entire library. One
> of the motives behind me writing my own also was to gain the learning
> experience. My library isn't as robust and well designed as
> boost::serialization, but for use cases where performance and archive
> size is of importance I think boost::serialization will find it hard
> to beat with its current design. Except for a smaller header which
> currently only keeps track of null pointers archive size is pretty
> much equal to that of a hand crafted solution.
>
> Citerar Robert Ramey <ramey_at_[hidden]>:
>
>> I'm convinced that the serialization library could be made to perform
>> significantly better.
>> To this end I've created a performance profiling framework which uses
>> boost.build
>> and library_status programs. As far as I know, profiling is only
>> implemented in
>> boost build for gcc.  Anyway,  I made the framework but haven't
>> found the time
>> to actually make the profiling tests.  These would highlight any
>> performance bottlenecks
>> in the library so that performance could be improved.  I've used this
>> approach
>> in the past on other projects with good success.
>>
>> If someone wanted to have a big impact on serialization library
>> users, he could make these tests and experiment with them.
>> Perhaps this would be a good Summer of Code Project.
>> Big impact - in a reasonable size project.
>>
>> I'm guessing that efforts invested in this activity would have
>> a much larger impact on boost users than trying to implement
>> a "lite" version of the library.
>>
>> Robert Ramey
>>
>>
>>
>> Sebastian.Karlsson_at_[hidden] wrote:
>>> Hi,
>>>
>>> I've created a small serialization library which takes a lot of
>>> inspiration from the great works of Robert Ramey. The main
>>> difference is that my implementation has favoured performance and
>>> file size while trying to maintain most of the goodies of
>>> boost::serialization. The largest thing missing, I believe, is
>>> pointer tracking. I'd like to make it available both for review and
>>> others in case there's any interest for such a library, but know of
>>> no good public place to make it available.
>>>
>>> // Sebastian Karlsson
>>
>>
>>
>> _______________________________________________
>> Boost-users mailing list
>> Boost-users_at_[hidden]
>> http://lists.boost.org/mailman/listinfo.cgi/boost-users