A method of recovery from a data storage system failure in a data storage
system having a host computer writing data to a first storage unit with a
first storage controller synchronously mirroring the data to a second
storage unit, and with a second storage controller asynchronously
mirroring the data to a third storage unit. The method begins with the
detection of a failure associated with the first storage unit. Upon
detection of the error or failure associated with the first storage unit,
the synchronous data mirroring relationship between the first storage
unit and the second storage unit is terminated and the host is directed
to write data updates directly to the second storage unit. Upon
correction of the failure associated with the first storage unit, the
asynchronous mirroring of data updates from the second storage unit to
the third storage unit is suspended and synchronous mirroring of the data
updates in a reverse direction, from the second storage unit to the first
storage unit, is commenced. When a full duplex state is reached between
the first storage unit and the second storage unit, the synchronous PPRC
relationship with the first storage volume mirroring data to the second
storage volume may be reestablished and host I/O writes to the first
storage unit may be resumed.