[xsd-users] Parallelization of the serializer

[Yury Zaytsev]

Hi,

I have found out that the runtime of my program that serializes a very 
large number of objects in a loop using the streaming serializer is 
entirely dominated by the calls to s.next(), that is at this point there 
is nothing I can really optimize at a higher level. Therefore, I would 
like to serialize in parallel from several threads.

This brings me to the question of what can I safely assume about the 
thread-safety of the serializer? A naive attempt to cilk_for my 
serialization loop has lead to a segfault, so I assume that it simply 
isn't thread-safe :-)

I realize that in theory, I can create a pool of serializers and spin on a 
lock in my worked threads until one becomes available, but this sounds 
like a substantial amount of work (as compared to a simple OpenMP or Cilk 
Plus solution).

So, as an alternative, would it be possible to make the serializer 
compatible with a Cilk Plus ostream reducer without too much investments?

https://www.cilkplus.org/tutorial-reducer-ostream

https://www.cilkplus.org/docs/doxygen/include-dir/group___reducers_ostream.html

Another approach could be to use Cilk Plus holders to store the 
serializers, but I haven't quite figured this out yet.

Have I overlooked anything? I would appreciate pointers to other 
approaches to parallelize with a minimal amount of extra work :-)

All the best,

-- 
Dr. Yury V. Zaytsev
Senior Software Developer
IT Data Production

TravelTainment GmbH
Carlo-Schmid-Straße 12
52146 Würselen/Aachen, Germany
yury.zaytsev at traveltainment.de

http://www.traveltainment.de

Amtsgericht Aachen, HRB 15873
Geschäftsführer: Bernhard Steffens (Vorsitzender),
Dr. Oliver Rengelshausen