Morning again guys!

I'm just running a test to see if using captainjacks suggestion will make a difference and how much.

I've reduced the processor useage to 25% which is running 16 threads, which is the way I'll go with the Captains suggestion, 4 clients, 4 hdd's 16 threads each. (while these threads are running randomly on the 4 cpu's, the hdd is where the test is)

My theroy is that if the hdd is the bottleneck, then this test should produce much reduced gap between elapsed and cpu times.

I've also cleared down all completed tasks and stopped new ones for now, and suspended network activity to allow me to view the tasks and their properties before uploading them..

I'll report back on the results in a bit, both with interim and completed.

Edit: corrected typo's!

It's interesting that you are ready to spend so much time on this ever lasting problem.
Because when you have tried everything possible to optimise your systems (which I think are already high end/top quality ones) maybe somebody will take two old posts of mine seriously.

My prognostic is that when you are short of new improvement ideas you will still be missing too many minutes of CPU time per WU.
Then take the time to read these two posts:
Most cores to crunch this project - Oct 25, 2011 8:48:08 AM
Most cores to crunch this project - Oct 26, 2011 11:41:28 PM

Then install BoincTasks (although it is a Windows program it runs very well under Wine in Ubuntu), make sure it updates frequently enough (say at least once per minute) and check yourself what I am saying in my posts.

BoincTasks will divert some more CPU time from the CEP2 tasks, but what you will see will be worth that cost.

I/O contention could explain that CPU time does not increase as expected (although several minutes is a bit too much in my opinion for modern disk units with plenty of cache memory), but it could never explain that CPU time goes backward.

But it is much better than in OP, isn't it?

Cheers

In your Original Post you mentioned that the difference is "1 - 1.5 hours".

So, if I calculated these "1 - 1.5 hours" against 18 hours (is it still the max for CEP2 task?), it is 92 - 94% utilization.

In the example of your current performance, the utilization is 96%, what I consider "much better"

Hm, that's pretty interesting. I saw your post/results and decided to check out my CEP2 efficiency as well.

I am using 4-cores of a i7-4771, 4GB RAM dedicated to BOINC, Mac OS X 10.9.2, and a 2TB 5400rpm HDD.

On 4 WUs at once, I'm getting 0.9686 CPU/elapsed time efficiency.
On 8 WUs at once, I was getting a horrible 0.8912 efficiency.

From your most recent 4 data points, it looks like you're getting 0.9583 efficiency. I'm not familiar with multi-CPU architecture, maybe you can try dialing the cores down even more to see if there is any efficiency difference. My theory would be maybe the OS/software or motherboard is not handling the processes efficiently.