Upstream work recently indicates that this may also be related to cache_swap_high garbage collection issues.
BUT we still need information about what HDD format has been used, and now also what traffic rates are happening at the time of overflow. Amos
