migrateBig_balancer.js runs the 'ping' command directly against the shards, and if that fails with a network error, assumes that that means the machine is having i/o problems and aborts the test. But aborting the test early doesn't really accomplish anything, as the test will still show up as failed in evergreen and has to be looked at, plus a single network disconnect does not necessarily mean there are serious i/o issues going on on the test machine. Instead of failing preemptively, the test should just run as normal and if there are i/o issues wait for the test to time out doing something, as all our other jstests do.