Well I am a refugee from WCG/CEP2, where eight Ivy Bridge cores would write over 1 TB/day. Nothing survives that. But Rosetta is not a problem at all for writes, so I have reset it to what I think are the defaults of 256 K cache size, with a 30 second latency. I then rebooted to clear everything out, and attached again to Rosetta. So we have a clean slate, and we will see how it goes.
https://boinc.bakerlab.org/rosetta/results.php?hostid=3399951

I worry a lot about "fairness" as perceived by the academic world. You hit a raw nerve with me. I am concerned that they don't make optimum use of the latest hardware (or software) to be "fair" to the older stuff. But whether Rosetta falls into that camp is not for me to say.

---

Well I am a refugee from WCG/CEP2, where eight Ivy Bridge cores would write over 1 TB/day. Nothing survives that. But Rosetta is not a problem at all for writes, so I have reset it to what I think are the defaults of 256 K cache size, with a 30 second latency. I then rebooted to clear everything out, and attached again to Rosetta. So we have a clean slate, and we will see how it goes.
https://boinc.bakerlab.org/rosetta/results.php?hostid=3399951

I worry a lot about "fairness" as perceived by the academic world. You hit a raw nerve with me. I am concerned that they don't make optimum use of the latest hardware (or software) to be "fair" to the older stuff. But whether Rosetta falls into that camp is not for me to say.

1TB/day?
Yikes!
The application has a bug.

The write endurance of SSD is about twice that of HDD. I have my BOINC directory located on the HDD in my system and have never seen that kind of write activity.

I have never been able to figure out the Rosetta credit system. IMO, any system that I cannot figure out is too complex.

The numbers are still low, so the cache size does not matter. Maybe it is just the work units themselves, or the strange way that BOINC credits are awarded? I don't really care about credits themselves, but I don't like using a machine inefficiently, and Rosetta does not give us any other way to measure output. So it looks like the Windows machine is more efficient insofar as I know, and I will just go with it, though I may change over from an Ivy Bridge to a Haswell (6 cores in either case).

Thanks for all the help and comments. It was very educational.

---