A very basic question

Anne & Lynn Wheeler

ref:

for a little more drift ... when the US hone datacenters were consolidated into a single center in the bayarea in the late 70s ... possibly the largest single-system cluster anywhere (at the time) was created. a variation of the "performance predictor" was used to monitor real time activity of all the members of the cluster and when a new branch office (or other sales, marketing or field) person initiated a login, the overall view of the cluster was used to decide which machine the user actually logged into (aka providing both availability and load-balancing at log-in)

In the 70s, the base system didn't have process migration between different processors in a cluster (once logged on).

Note, however, one of the time-sharing service bureaus using the same (cp-cms) platform, had done such a enhancement in the mid-70s ...

this particular service bureau had moved to 7x24 operation with customers world-wide ... and so there was no period where the system was totally idle and could be taken down for service. When a member of their (mainframe) cluster configuration needed to be taken offline for service or preventive maintenance ... it was possible to migrate all active processes off the piece of hardware (to other members in the cluster, transparently to the end user), and take the hardware offline.

they had datacenter near boston and another in san francisco and claimed to be able to do transparent process migration between datacenters (as well as between members of the cluster within a datacenter) ... modulo having replicated data at the two centers.