After moving nodes of the zewura SMP cluster (renamed to zebra1-12) to the new computer room some of the nodes appeare to exhibit very rare memory write failures under very intesive memory stress test. The problem is not reproducible, it occured only few times during several days of testing. We consider it almost impossible to occure in normal operation. The problem was reported to the supplier's technical support for futher detailed diagnostics.
Nodes are being returned to the normal operation. Despite the problems are not expected, we kindly ask the users for reporting any suspicious behaviour.
We apologize for any inconvenience caused.
Ivana Krenkova
MetaCentrum & CERIT-SC.