I can see larger amounts of urgent writes, see here:
https://www.dropbox.com/s/a68a307eu12tq ... 7.png?dl=0
I changed the setting for defer-write to 10s to see if thats better.
I will consider using individual cache space. If using individual cache space, will blocks that have just been written and are still existing in write cache space be available for cached reading if requested shortly after?
Consider rendering an image sequence and reviewing it immediately after rendering. Thats 90% of the workload here.
You did not answer this question yet:
When outstanding write cache data is flushed to disk, will PrimoCache then automatically free up that cache block immediately or will it still be present until another read request pulls it or new write cache data might need to overwrite it, on a least used - first overwritten policy?
I think that would make a lot of sense to keep write cache data even AFTER flush to disk operation as long as possible for potential read requests...
You can also see that I have L1 enabled and that it helps mostly on deferred writes. This volume is not yet in use for image sequence rendering, currently being used as a backup mirror. I will manage to use it as rendering drive early next week, to get a better idea of daily use scenario.
Here some numbers from real work load drives, both v2.4 server:
https://www.dropbox.com/s/d3te4606aqdrr ... 7.png?dl=0
https://www.dropbox.com/s/silra59uwepi7 ... 7.png?dl=0
>11 million trimmed blocks...
and the new mirror drive with 3.0.2 beta server:
https://www.dropbox.com/s/a68a307eu12tq ... 7.png?dl=0
Thanks for your support.