• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Rosetta memory leak on 4p??

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

donuts

Member
Joined
Dec 5, 2006
Location
SC (now)
I've had this problem with Rosie on my 4p. CPU usage will drop to 60-70% after running a while. (hours)
Doesn't appear to be a temperature problem, a reboot will kick it back up to 100% usage.
This has been an issue for about a month and only on the 4p.
I am running the latest BOINC manager and an old optimized client for 4p rigs (Tear's)

I'm thinking memory leak as I have observed memory usage in the high 40gb usage, and a fresh reboot will load memory in the mid 20gb range. When it's in the high 40gb usage, cpu usage drops.(system monitor in Ubuntu reports this, Ubuntu version of "task manager") System currently has 60gb installed and BOINC is allowed 95%.

Any thoughts on how to fix this? Reboots every 12 hrs can be a pain when this rig used to run for weeks on end.

Heatsinks have recently been cleaned out and board is on a Spotswood tech tray, so no case.
Set the client to run out cached units, may try a newer version of the client. I'll also try some memtest runs later, just to rule that out.
Don't really think it's the client that's running though, my 2p is also running that client fine.

edit:
Memtest may be put on hold for a few. 60gb takes a while to test. 10 hrs and I haven't completed the first pass. Still running, no errors so far.

edit#2: successfully completed one pass. 13hrs. back to clearing cache.
 
Last edited:

Evilsizer

Senior Forum Spammer
Joined
Jun 6, 2002
have you checked on the bonic forums if their is issues with amd 4p systems? wonder if it might be something they havent seen yet they can fix or the possibility something else is eating up ram?
 
OP
donuts

donuts

Member
Joined
Dec 5, 2006
Location
SC (now)
I was blaming Rosetta as it's pretty much the only thing installed. HDD and Ubuntu install is only a couple weeks old. Google search turned up a couple hits, and only specific Rosetta units. Nothing new, old posts. That said, it may NOT be Rosetta per say, but a Ubuntu update that broke it.
I'm clearing out the Rosie cache now, I'll deal with this in a couple days. If I can get a couple good passes of memtest complete, then I'll probably do a fresh install with NO updates and see how it works.

The client is an old modified client that was done by a 3rd party, thus no official support for it. (modifications are to keep all the 4p cores fed) Maybe try the new "official" client to see if the old issues were ever fixed... The 4p is a new platform to me, not new to DC projects.
Folding forum has an old sticky thread on 4p rigs when SMP was the big ticket.

Prior to Christmas, this rig ran fine. If I can rule out hardware failure then I have to look at software.

9:00 pm edit.
Further narrowing down the problem, quite a few "minirosetta" units still holding on to 1.5-3gb of memory AFTER pausing the client. Nothing is suspending to ram in the client settings. My gut says an update along the way broke it. I seem to recall some updates to the BOINC client when I reloaded the operating system. These updates were through Ubuntu update, not the BOINC client.
Another note, unfortunately my lack of cache seems to have fixed itself after months. (see couple threads down). Instead of 12+ hrs like I used to have, now I have DAYS worth.
 
Last edited:
OP
donuts

donuts

Member
Joined
Dec 5, 2006
Location
SC (now)
Back up and running. Ran overnight and it didn't seem to "hog" memory, knock wood.
Memtest all day yesterday (whoopee, 1 complete pass...) Reloaded OS with updates BEFORE I installed BOINC. Hopefully this will fix it.
Ignoring any BOINC updates for this machine. We'll see if that works... Crosses fingers..

Couple days later, ram usage holding steady. Seems like it WAS a Ubuntu BOINC update that broke it. I will continue to monitor it. Seems like I also figured out the work buffer. For future reference, there is a minimum setting also. Default is 0, set it for a day and I have plenty. How I missed it after so many years...
In my defense though, usually the max buffer will load up the rigs.
 
Last edited: