[StillBored]

100% agree, but AFAIK the HW isn't doing lockstep on modern zseries. The nonstop was also HW lockstep many years ago and they converted to a checkpoint restart model, which AFAIK is the same on zos/etc with help from the "HW" which is just "software" running in the LPAR/etc.

Regardless, there have been various clustering/SSI/etc software layers designed to make windows/linux/etc behave as though it were lock stepped as well via software/hypervisor checkpoint restart.

So it is not impossible, but you don't get it out of the box because most of these applications have moved the fault tolerance into an application + database transactional model where the higher level operations aren't completed until there is a positive acknowledgment that a transaction is committed (and the DB configured for replication, logging, whatever if needed, blocks the commit until its done).

So, yes that model requires more cognitive overhead for the developer than Cobol batch jobs, which tend to be individually straightforward (or the ones i've seen the complexity is how they interact). But the results can be the same without all the fancy HW/OS layers if the DB is clustered/etc.