All my machines run Ubuntu. Some 64bit some 32bit. These issues are confined to the single core machines. The other 32bit machine is a multi-core and had no trouble.

Another data point: I've had one WU with RC=0x4 marked as valid!

The wingman was slower and didn't make it to job #4 at all!

Makes me more curious about the results validation side. Maybe a result with RC=0x4 is actually ok, but the validator isn't happy with it when another, longer, result doesn't include that? Alternatively, maybe a result with RC=0x4 is not ok, but gets passed by the validator anyhow when the wingman is slower?

Or even better, quit trying to force quotas via time limits.

Well, so far WCG is not budging from limiting the upper run times and as we progress into the research and average toughness of the tasks will increase to the point where quads too will need 12 hours, don't think it to happen that P3/P4's are allowed to run all 16 job within a task to the end over several days [let alone the part-time devices], with checkpoints that are multiple hours apart (can't do anything about that unless crippling the system for users and start saving 100s of megabyte chunks to disk at short intervals). Then on each boot it will step back those hours without LAIM on (recommended for this project), skipping back to last checkpoint when the user has set the client to pause when in use. Need to have a hard think about that, without knowing all the facts for consideration available.