Comment by maverwa

That’s timing the cache, that’s old stuff by know. As I understand, this writes a relatively large file („Gigabytes“) using this OPFS api, which is different from the „localStorage“ api. This seems to use actual filesystem storage on the client, instead of living completely in memory (which may be reasonable given the size of files supported). This allows to actually time SSD IOPS latency by doing random reads.

Collected enough of these samples, together with the information of what else runs on the host, put that in the ML-Blender and the result will be able to tell you, with some accuracy, from a given set of samples, what’s running on the host.

I am sure i misunderstood some things because there are so many caches and unknowns in that setup that I struggle to understand how there could be any correlation, but that’s my understanding so far.