Notes
 -----

  For now, ssh tests assume REMOTEHOST is set to a ssh server,
  otherwise localhost is used.

  You must have sshfuse installed (for file tests) and ssh keys
  in their default location.


 Performance
 -----------

  The ssh file adaptor accepts local file:// and remote ssh://
  based URLs, and, of course, any:// based URLs.  All ops which
  end up on the local file system are transparently performed,
  all operations which end up on non-ssh remote systems are
  eventually declined.

  The adaptor performs *all* operations by internally calling
  SAGA, so it in fact only translates URLs back and forth, and
  invokes SAGA on these translated URLs (*).  That architecture
  comes with an intrinsic overhead, and effectively doubles the
  SAGA intrinsic latencies.  In most cases that should not
  matter compared to the netowrk latencies involved, but in rare
  cases it can add up.  For example, consider a slow remote
  adaptor which needs time to decline an operation - that
  latencu may occure once on the top SAGA level, and once more
  inside the SAGA op called by the ssh adaptor.


 Fixes for Performance Issues
 ----------------------------

  A possible fix is to equip the local file adaptor with a local
  context, and have the ssh adaptor use a session which only
  contains that local context, and the ssh contexts.  This
  should effectively limit the number of adaptors used inside
  the ssh adaptor ops.  

  Another option is to move all inter-adaptor operations on impl
  package level (see discussion on list), which would allow the
  ssh adaptor to more aggresively filter for ssh:// based URLs
  only.

  Other than that, there is not much more one can do, apart from
  educating users that in particular the use of any:// based
  URLs *can* come with a significant local performance overhead.



(*) Well, of course the adaptor does more, such as mounting
    remote ssh file systems for example.  But you got the idea
    I hope.