Hello, do you know what is slowing down scrubling the time line when i load sequence of particles from prt files with 7 milion particles, because the same sequence but in bin format is realy much faster also in 100 percent of showing
PRT files are compressed, so there is an unzipping step involved which is unfortunately running on a single thread per sequence right now. I suspect that it would be slower to unzip 7MP than to read them from an uncompressed binary file. Did you use the PRT Loader to test the BIN file, or something else?
yes the bin files are loaded in prt loader and it is realy huge difference. Do you plan any multi-threaded unzipping or different method of multi-treaded compression? Thanks s lot
Currently PRT Loaders that specify multiple sequences in its file list will load the individual sequences in parallel. The current PRT format does not allow for concurrent reading, so that is one thing we are considering for the next spec.
I didnot think concurent reading, i had separate prt with bin sequence and separate for prt sequence, and the speed is realy good with bin files, not with prts
What Darcy is saying is that if you take the BIN file and save 10 PRT files with 1/10th of the original sequence (using Magma deleting all particles but every Nth with offset of 1, e.g. first sequence containing particles 1,11,21…, second containing 2,12,22… etc), the loading of these 10 PRT files in a single PRT Loader on a system with multiple cores and esp. from a very fast hard drive (say, an SSD) can be up to 10 times faster. The PRT Loader will create one thread for every sequence specified in its list and the more cores you have, the faster it will load. At Siggraph this year, we were loading 5 million particles using 10 sequences on a machine with only 4 cores in 0.7 seconds from a FusionIO card. Unfortunately, it is up to you to create the multiple sequences (usually partitions, but you can split one file into many as described above)…