You could both be right -- could be that a hub/spoke model would fare better in a software failure, because everyone would return to the hub and things would be straightened out the old-fashioned way.