we are having an increased number of IO errors here lately, which is intensely impacting our output rate right now. A lot of times, the following errors pop up:
particle_file_istream_factory: The input file “Z:\project\splash.0123.prt” does not exist.
or win32_reader::open() CreateFile() failed: Access is denied.
or prt_read_interface::open() CreateFile() failed: Access is denied.
or error:…getOverlappedResults() [<sorry, don’t have the exact error message at hand right now)
Surely, the very prt-file does exist and is definitely accessible. Also, the affected frames might fail with the aforementioned error once or twice, but will eventually render fine. Our server logs show no sign of any declined request or irregularity at the highest verbosity level.
These erros are happening both with rendering Krakatoa AND just loading prt-sequences to feed into Frost.
I have a gut feeling this might all be caused by similar things? Would it have to do with multi-threaded loading? And if so, how would multi-threaded loading be turned off in non-Krakatoa-Rendering mode?
All these errors are usually symptoms of network hickups.
Multi-threaded loading is now disabled by default since v2.3 of Krakatoa MX, mostly because your company requested it a few years ago after a similar problem was reported. We also exposed the switch IN RED in the UI so nobody can leave it on by accident. And we turned it off completely in Krakatoa MY and C4D without even an option to turn it on.
How many machines are reading PRTs at the same time when that happens? And how many other machines are using the network for I/O during that time (e.g. loading huge textures in V-Ray, for example)?
yeah I remember back then we were able to fix our issues by disabling multi-threading completely. And yet, as far as I remember back in the days we ‘simply’ had performanec issues, now errors. Would the likelyhood of the errors described increase with each node loading multi-threaded still?
As for the situation:
From a linux-based file-server (which is showing very low load)
20 machines constantly pulling prts for Krak
40-60 I’d say pulling for Frost (would it make a difference actually, for Krak vs. for Frost?)
10 writing prts
Overall, I’d say 200 nodes doing other stuff throughout the network, but not working on the drive holding the prts.
I checked with the Frost developer and it is single-threaded when loading particles via its own built-in file list.
If you are loading particles using a PRT Loader, then the actual loading might depend on the Krakatoa settings.
Since Multi-Threaded loading is off by default in KMX v2.3 (the option is called “>Fast PRT Loading” in the Main Controls and turns RED when on), chances are you are not loading in parallel anyway.
Unfortunately, I am not a specialist when it comes to networks and hardware. Can you please email support@ so you can discuss this with people that actually know that matter?