Use an HDD instead of an SSD if you're going to do CEP2. Any write speed advantage a new SSD has compared to a short stroked HDD, is more or less gone when the SSD fills up and begins to overwrite old data. SSDs are not ready for write-intensive tasks or virtualization, at least not until XPoint is paired with NAND in future SSDs, sometimes late next year.


I'm crunching 12 threads OET in an Ubuntu guest located on the first HDD (C:) and 12 threads CEP2 in a W7 host on a second HDD, without any problems. Since I put the BOINC Data file on a second HDD and short stroked that HDD down to 2,5%, the previous problem I had with resetting CEP2 wus is now gone. The computer needs 10W more now, compared to when I only had one HDD and did 24 threads OET, but that's a small price to pay for a greener planet. Let's just hope Bao Group gears the algorithms towards finding low resistance, carbon based molecules with band gaps suitable for photovoltaic cells, instead of carbon based LEDs. She is after all sitting on LG's science advisory board--the biggest OLED manufacturer on the planet--so there's room for doubt ...

A quick notice of what I'm seeing, after allowing BOINC to use up prety much all my main memory. WCG CEP2 executable:

- uses 100 - 110 MB accordsing to Task Manager
- has a working set of 100 - 110 MB according to Ressource Monitor
- shows disk activity every few seconds, often in the 0.x MB/s range but also frequently about 6 MB/s for a few seconds
- SSDReady shows projected 160 GB/day (normal would be 33 for my system

4.3 GB of 8 GB main memory are used. During the observation time the WU went from ~4% to 6% - not sure if things change later on. The rest of the config is: Win 8.1 64 & BOINC 7.6.6 (not sure what else might matter).

For now I would conclude: "I don't think I'm seeing the behaviour I should have seen according to your comments". And "if CEP2 wants more memory, it's not getting it or not taking it".

MrS

Would putting the BOINC data files on a USB 3.0 thumb drive help?

I have done this thrice, not only BOINC, but the OS (Linux Mint) also. The first time I used the cheapest no name drive I could find, I think it was $3.99 for an 8 gig drive. It lasted about 4 months and then froze up. The second time I used a slightly better drive, $5.99 for a 16g (PNY) drive( prices have come down lately). That one lasted about 10 months and then froze with the message it had turned into a read only drive. Currently I have another 16g drive (SanDisk) running (about 2 months). They have been run on a Dell 2650 server with 8 cores running 24/7. All have been USB2.0. I am quite sure they can't use much electricity. Even though they fail eventually, I am satisfied with the performance. Been running mainly MCM1 over this time period.

Cheers

TBMS, your data is solid. I only disagree at one point:

The writing performed there is constantly hammering the drive with 4k writes, at the maximum performance the drive can sustain. There's not a single break for the controller to recover. That's not what I see during CEP2. It often writes, yes, but that's a small amount of data every few seconds. And then a few 10 MB maybe once a minute (not measured). There is a lot of idle time in between, hence any half-decent SSD (not the JMicron from 2009) won't have a problem with that load. Except that it has to "pay" for the writes, of course.

An some more updates:
A WU has finished with the setting of more available RAM. It increased credits per hour by a factor of 2 compared to the previous one! I don't know how statistically valid those numbers are, though (just a single sample for each setting).

The current one is between 24 and 26% and still shows 110-120 MB usage in task manager, whereas the working set has increased to 210 - 220 MB. Overall memory load still 54%, with 80% allowed for BOINC.

MrS