Maybe an automated app_config should be loaded from the CEP2 science...so that it runs max 2 apps concurrently...or is that too much for the WCG / CEP2 team? ;)

How do the disk drives compare on the two computers? With the very high write-rates of CEP2 (don't know about FAAH), it is quite possible that you have write-contention when accessing the drive. Some SSDs for example are rather sensitive to this, and can't take high write loads. It depends on how much cache they have, etc.

On all the machines where I run CEP2, I use either a ramdisk (Primo Ramdisk or Dataram RAMdisk), or else a write cache (PrimoCache), and basically don't get errors.

For using a ramdisk, you first create one of large enough size, which depends on how many CEP2 work units you want to run at once; about 1.5 GB per work unit is about right, and then install BOINC so as to place the BOINC Data Folder on the ramdisk. Then all the reads and writes are to main memory, which is much faster than any drive. Setting up a write-cache is somewhat simpler, since you don't have to change the BOINC data location, just create the cache; a couple of gigabytes should do it, though I use more. Then the writes all go into main memory, and the old data is flushed out to the disk drive when the cache fills up. It will solve any problems due to high write rates.

I don't know if this counts as the same problem, but I too have Windows 7 Pro and when I simultaneously run 12 threads CEP2 and 12 threads UGM, some threads, usually CEP2, will reset, sometimes all the way down to zero. It's been like that since I bought these used computers and I'm loosing on average 5-10% of computing time. Many kWh down the drain, every month.

For a while I used to manually start the wus after each involuntary resetting, one by one, with one minute in-between them, but now I'm tired of doing that. Now I'm just counting on, hoping, that some of the wus will have different crunching times and therefore achieve a finishing time that's equally spread between the wus.

I suspect this problem exists because the server sometimes doesn't allow immediate upload, so some type of memory instability occurs in my computers when wus are waiting to be uploaded. I have 24GB registered ECC RAM and an HDD.

Or, maybe it's because the chipset is running at 67C? I've ordered a second 40mm fan to each of them.

I could probably mitigate the problem by crunching fewer CEP2 wus, but I don't want that. CEP2 is the main reason I'm on WCG. Maybe the above mentioned write-cache could solve the problem? Thanks Jim1348. I have enough free GBs of RAM for a write-cache but not for a RAM-disk, at least not while the hyper threading is turned on, since CEP2 requires 2GB per wu, according to WCG . The hyper threading stays turned on, but as it stands now, the 10-15% performance gain derived from the hyper threading is almost entirely lost to the involuntary resetting. A write-cache, such as Supercache Express or PrimoCache , could be the answer. The former supports NUMA, which could perhaps be a better solution in a dual board.

Each of my computers are writing about 13-15MB/s, which amounts to about 440TB/year. I need 12GB RAM for CEP2 and another 1,2GB for UGM, which comes to 13GB. Another 4GB for the OS, and I'll have 7GB over for the write-cache. It takes a 7GB write-cache 500 seconds, or little over 8 minutes, to fill. Will this be the magic bullet? Only if the HDD can write all the data without the cache filling up.


Edit:

After installing a test version of Supercache, I can now say that it doesn't work. BOINC is pretty much ignoring Supercache and continues to write to the HDD at the same rate as before. Almost nothing is being read from the cache. So I'll uninstall it.

: (

I've already lowered the number of CEP2 threads, but you're probably right. I'll lower them to 10, but if I could, I would run CEP2 on all 24 threads.