Hi Sebastian,
yes the machine is quite a bit loaded when these copies are running, but it's also rather beefy - 12 real cores and 60GiB of RAM may sound excessive for a backup server (CM and MA), but making SOS fly has a price. SO Appliances come with 72GiB RAM or more for a reason. I hesitate to reduce my parallelism (backups go to 8 B2D streams per-object, and those are copied to tape with a multiplex concurrency of 8) as it worked well so far (started at 9.02 and worked right through to 9.05 which was quite a nice stabilization and bugfix release - I upgraded to 9.06 mostly because of QCCR2A62437 [SOS Housekeeping stops working after a week or two]).
The situation seems to even worsen: Today I was greated by a copy job that reproducibly fails to read a certain B2D medium with Cannot read from device (JSONizer error: Read failure). A reboot of the entire machine and omnidbcheck -extended didn't change that behavior. The B2D medium turns poor when this error strikes, but it verifies good (turning green again, yet the entire pool somehow doesn't pick up that change and stays poor/red)...
I'll monitor this situation closely for the next days and if it doesn't turn out to be a glitch that goes away as mysteriously as it came (which wouldn't be a first), I'll have to punt this to support. Let's see how things develop.
Thanks,
Andre.