Lol, modded the app_config to 3 ugm 5 mcm and now ugm runs more efficient than mcm, albeit it's at the 2nd decimal level . All now solidly into the >= 99.4 percent efficiency realm. Note, still running with a write to disk setting of 1000 seconds!

Now that data_dir is no longer indexed, the slide file timestamp changing, used per docs by the indexer to revisit the files, has possibly become a quasi moot issue too. It is slowly becoming evidential that ugm was setting something off on the i/o front, no explanation why 'mcm only' gave no performance issues.

Now we also could use a random checkpoint offset time for each task, so if 8 resume or start simultaneous, they do not sync-checkpoint at fixed intervals, which is what they're now doing.

Momentarily considering if this also impacted cep2 that never ran that great on this node. Something else to test.

Finally, hoping with 3:5 crunching in present config the feeders keep up the ugm supply, else it is 'simap we have supply needs'.

I have not looked at the part of the client code in detail but I believe it has something to do with the client checking if any of the files need to be deleted.

Thanks,
armstrdj