View Revisions: Issue #33

Summary 0000033: New high-efficiency parallel stacking algorithm
Revision 2016-10-16 00:54 by vinvin
Description Stacking is already parallel, but done in a quite simple manner, with OpenMP parallelizing the main loop. A more efficient and smarter way could be done, using a hand-made thread that produces data and leaving the OpenMP threads consuming them.
The data producer would have for only task to read files from the disk. We should not do this concurrently as we do today, because it's slower on hard disk drives and makes them work too much, and it won't go faster anyway on ram or flash disks.
The consumers would use this data from a fifo, working only on userland CPU algorithms.

That will not be done in the near future and in the future there may not be rotating disks anymore so that will be as good as useless.
Revision 2015-07-08 11:35 by vinvin
Description Stacking is already parallel, but done in a quite simple manner, with OpenMP parallelizing the main loop. A more efficient and smarter way could be done, using a hand-made thread that produces data and leaving the OpenMP threads consuming them.
The data producer would have for only task to read files from the disk. We should not do this concurrently as we do today, because it's slower on hard disk drives and makes them work too much, and it won't go faster anyway on ram or flash disks.
The consumers would use this data from a fifo, working only on userland CPU algorithms.

That will not be done in the near future.