Today we started to use CTA at a high load for the first time and the taped immediately crashed. We tracked it down to a single taped process for a drive using 24+ GB and we have two drives on a node with 32 GB. So things are stable with one drive, unusable with two per node.
Are there tuning settings we can turn down which would allow us to halve the memory usage? If it matters, we are writing 4 GB files.
At RAL we use a combination of buffer size and buffer count to limit the amount of memory each taped can use (in cta-taped.conf) -
# BufferSizeBytes
taped BufferCount 5120
# BufferSizeBytes set in aquilon from old CASTOR default
taped BufferSizeBytes 5242880
So if you allocate 30GB for the taped’s, thats 15GB for each drive/daemon, so assuming you stick with the default buffer size, your count would be 3072 (if I have my maths correct?)
Tim