Is the write_back option the same as the buffered option mentioned here:
http://linux-iscsi.org/wiki/LIO
It says the buffered option is unsafe and not to use it.
I created a 10TB file as a test LUN yesterday on a ZFS dataset and I enabled
write_cache and emulate_write_cache=1. Performance was staggeringly good
compared to using a ZVOL. The target LUN is shared to multiple initiators
using Tiger Store software and it is formatted with NTFS.
Does write_back and emulate_write_cache do the same thing?
I'm going to moving to using a file instead of a ZVOL for production. I'm
want to set this up safely though and to be able to recover from a kernel
crash or something similar. The servers are on UPS and they are all using ECC
memory. I don't mind that having to do a NTFS disk check if something
unforeseen happens, but having a whole corrupted LUN is not acceptable.
Any advice would be greatly appreciated.
Chris