Hello again, and sorry for causing this confusion with my short reply!

So, there are different definitions of "Multipath I/O" on
the mainframe:
- the ESCON/FICON multipathing which allows concurrent
  access to the same control unit (eg. a shark) which is handled
  in the architecture and which works with Linux as well and which
  is sufficient in most scenariosn (eg. different OS-images doing traffic
  to various devices concurrently)
- the PAV multipathing which makes the same device show up
  multiple times in linux (multiple dasd devices correspond to the
  same real device). This allows concurrent I/Os to the same
  device, if an utility can detect that these device nodes correspond
  with the same device and combine them to do load balancing.
  As such utility, one can use LVM with the LVM-MP patches or
  the EVMS project which I did not try myself yet but is heared to be
  a very nice tool which is a strong candidate to replace LVM in
  the kernel. These utilities do also apply for SCSI-FCP multipathing.

I do agree that the striping workaround I suggested is _not_ a
replacement for a PAV soloution, which offers both better reliability as
well as better performance than the workaround.
The workaround however can help to boost access to a single volume
with todays Linux distributions and without need to patch and rebuild
anything because it helps to utilize different ESCON/FICON pathes in
such a scenario.

mit freundlichem Gru� / with kind regards
Carsten Otte

Reply via email to