Got one that might be erroring out. Info dump below.

Woke up this morning to find BETAs had gone out. I got 5. 4 on Linux, one on Windows 10. Four so far have completed, with times between .02 (1 min 12 seconds) and .85 hours. All have 49 checkpoints in the log file. I noticed on two workunits that there is some pretty large variability between machines though. One of my wingmen so far have approximately the same time spent as I do (.02 and .02 on unit -1292). Two other wingmen have different times than I did (on unit -0203, I spent 15 mins on it while they spent about 5; on unit -1109, I spent 51 minutes but they spent almost 3 hours).

One wingman has not yet replied to a very short unit (.03 hrs), but they could be working through a large cache, perhaps?

There is one that seems to be an outlier though. Unit -0392 has either started looping, or has slowed down to a crawl. Currently has 12 hrs 30 minutes, and it's at 76.001% completion. It seems to increase by .001% approximately every 90-120 seconds, which means it's slowed significantly. I think I saw it get up to 76.007% before returning back to 76.000%. The "time since checkpoint" also does not seem to be accurate. When I first checked, it was 10 mins since checkpoint; now after the loop it is 6 minutes. If it loops again, I will try rebooting the system (so much for my ARP tasks lol) and see if that assists the unit to completion. Finally, wingman has also not reported in on it.

System in question is a dual EPYC 7601, 128 gb ram. Currently crunching 29 ARP tasks, 1 OPN task, 4 HSTB tasks, and around 94 SCC tasks. (I believe that at least a couple MCM task has also been crunched in the past 12 hours or so.) This system is also the same system that crunched -0203 in 15 mins. While time to crunch units can vary significantly depending on the unit (had some ARP tasks as long as 2 days, and HSTB as long as a day), it's rare that a unit takes 2x the norm to crunch.

I'll report back in a couple of hours.

Both WUs that I got on a remote system have been returned (1.22h & 0.65h CPU time), so did the wingmans. Now they are sitting for almost a day in PVa jail...

Ralf

Update to my previous report:

I was slightly mistaken, but figured out what was happening. The one final work unit was not looping, but rather progressing strangely. It would progress .001% about every 2 minutes, but once it would tick over to XX.010%, it would instantly advance to the next checkpoint, and add 2%. So, it would go from 74.009% to 76%. I misremembered the 74.009 as 76.009, which gave it the appearance of looping. But what was actually happening was each 0.001% increase was actually a 0.2% increase. So, it seems that there were 500 steps in that work unit?

Overall, it finished with just under 16 hours. Still waiting on wingman.

Yes I am aware that we can set how often they are written. The data needed to make checkpoint must still need to be written to the hard disk even if we have preferences set for example for 17 minutes. Will all tasks write 49 checkpoints regardless of length?

Uplinger thanks for your response. Looking forward to helping out this project whether it be in beta or when it is in production