The frame store snapshots made it fast to load large frames stores, but it was slow to freeze because the GC of the large global store. I have changed the snapshot format so it can now load the snapshot into "frozen" heaps. These are then read-only and all the objects have the mark bit set, so now GC takes less than 0.5 seconds.
I had to change the object layout in the frame store. One benefit of this is that a store can now have twice as many objects (512M to 1024M objects).
I have also implemented support for reading frames in RDF Turtle format, which we will need for Wikidata lexicographic data.
The frame store snapshots made it fast to load large frames stores, but it was slow to freeze because the GC of the large global store. I have changed the snapshot format so it can now load the snapshot into "frozen" heaps. These are then read-only and all the objects have the mark bit set, so now GC takes less than 0.5 seconds.
I had to change the object layout in the frame store. One benefit of this is that a store can now have twice as many objects (512M to 1024M objects).
I have also implemented support for reading frames in RDF Turtle format, which we will need for Wikidata lexicographic data.