Using what amounts to an I/O logging system to which you write data synchronously vastly reduces the cost, but that still has to be placed somewhere and with appropriate network communications such that it, and the onward asynchronous copying, is not affected by the disaster. A wide enough problem could affect all terrestrial communications to I guess that satellite might be the ultimate fallback.

In the case of the vast majority of our systems we elect for asynchronous replication on cost and performance grounds and work on the basis that the resulting inconsistancies can be sorted out by manual and exeception processes (and this is truly for disaster scenarious – not as an HA approach). However, that’s not an option with many safety or regulatory related systems. For instance, losing a few bank transactions isn’t an option nor would losing critical health care data.

If this sort of approach works, and it gets rid of the logical issues of recovering from inconsistencies caused by asynchronous replication then it could be very valuable indeed. But DR is an incredibly difficult thing to get right.