Two questions:
(1) What level of memory use are you seeing for rosetta@home (with graphics) on your computers?
-Well with the Windows XP SP2 Stripped OS That is currently in developement and is currently out for Public usage and Beta testing. the Average Ram Usage is usually under 256Mb
If we Invoke the Bench Mode we can get it down under 128MB
(2) Should we disable the graphics in the next release to reduce memory use for larger proteins?
(at ~100Mb per work unit, there should be no problems on most machines).
-Actually I think you should give the option of it but outside of that I see no real need to remove it.. but then again this is my opinion.
____________
How come since version 5.12 appeared all my movies start very slowly with WMP (or Media Player Classic) using ffdshow codec? I have two systems running Rosetta and one had a 5.12 project and showed this problem. The other one was still working on a 5.07 project and played movies fine. Now this other system works on a 5.12 project and movies load slow.
It's not a processor issue since if I suspend all work, and no processor is used, movies still take some time to load. Problem dissapears when I close the BOINC manager.
And I'm not very sure if it's connected to ffdshow, I seem to have the problem with gamespot's movies that are encoded with wmv codecs and I don't use ffdshow to decode those.
well now, there have already been two reports of this type of activity reported in relation to 5.12 in Ralph. I've seen something that I'm not willing to pin to Ralph/5.12 just yet, but this thread might get interesting.
Two questions:
(1) What level of memory use are you seeing for rosetta@home (with graphics) on your computers?
Under Windows, Rosetta used to take 60MB to 160MB of memory (with some HUGE WUs which needed up to 250MB a month ago) and 300-450MB of virtual, until recently. Latest v5.12 seems to have drastically reduced memory to "just" 60-75MB and 300-320MB virtual.
For monitoring processes under Windows I recommend the excellent free Process Explorer util.
Oddly, under Linux, Rosetta takes about HALF of that (30-40MB and up to 85MB).
This is ofcourse empirical, from casual observation.
ps u -U boinc
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
boinc 2120 0.0 0.5 7468 3756 ? S Apr27 0:17 ./boinc_client
boinc 4763 10.1 11.0 116472 82112 ? SN May09 118:20 rosetta_beta_5.12
boinc 4764 0.0 11.0 116472 82112 ? SN May09 0:00 rosetta_beta_5.12
boinc 4765 0.0 11.0 116472 82112 ? SN May09 0:00 rosetta_beta_5.12
boinc 19053 69.3 4.0 94640 30264 ? RN 13:43 182:12 rosetta_5.12_i686
boinc 19054 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
boinc 19055 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
boinc 19056 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
(2) Should we disable the graphics in the next release to reduce memory use for larger proteins? (at ~100Mb per work unit, there should be no problems on most machines).
Please don't disable graphics. Remember that for many crunchers, the "reward" is the cool screensaver graphics and the credits.
Please don't disable graphics. Remember that for many crunchers, the "reward" is the cool screensaver graphics and the credits.
If all else fails, perhaps you could add a note somewhere that Graphics may slow down some systems, and in that case, the project suggests turning the graphics "off" [or whatever is the appropriate language].
This would at least give people a heads-up to potential memory issues. I totally agree that the decision to turn off the graphics should be up to the individual user.
I've been trying to track down memory usage (Windows). I often find it seems some specific WUs, while saying their memory space is modest (150MB or so) that they generate PILES of page faults. On the order of 1,000 per second for the entire run of the WU (24hrs in my case).
I haven't been able to track down whether it's due to the protein, or the method of analysis of a given WU. In fact, I think I've had two running at the same time on my dual core CPU and both for the same protein name, but different methods, and one generates 1,000+ page faults per second and the other virtually none. I'm generally monitoring the page faults while NOT viewing the graphics.
Using less memory is great. I agree though, don't change the project stance on memory reqs prematurely. A note (perhaps even within the new protein description) that viewing the screensaver or graphic may slow performance on some systems would be a good idea.
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
In general, running a BOINC project is going to use memory on your PC. And with that memory in use, it will often take longer to load any other application. Video needs both the application and the video data, which requires lots of memory. And if BOINC continues running while computer is in use, it will continue to compete for memory, and thus further degrade startup of other applications due to the memory contention.
To investigate further you might try suspending boinc (Commands tab, "Suspend"), then verify in task manager that the crunching threads are no longer using CPU time... then start the video.
The fact that this is a new problem for you with 5.12 would seem to disprove the above theory though :(
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
I've been trying to track down memory usage (Windows). I often find it seems some specific WUs, while saying their memory space is modest (150MB or so) that they generate PILES of page faults. On the order of 1,000 per second for the entire run of the WU (24hrs in my case).
I haven't been able to track down whether it's due to the protein, or the method of analysis of a given WU. In fact, I think I've had two running at the same time on my dual core CPU and both for the same protein name, but different methods, and one generates 1,000+ page faults per second and the other virtually none. I'm generally monitoring the page faults while NOT viewing the graphics.
Feet1st, right on, I just checked and I find the exact same behaviour. On a PC with 1GB of memory and very little else running besides Rosetta.
One Rosetta process has page fault delta of 0, the other between 1000-2000/sec
In my case, the one doing 1000+ pf/sec was JUMP_CLOSE_CHAINBREAK_ALLBARCODE_1a7aA_SAVE_ALL_OUT_493_3201
once suspended, overall system pagefaults/sec went down to almost zero. Resuming the WU now (having changed nothing) the overall pf/sec delta is close to zero.
Feet1st wrote: "To investigate further you might try suspending boinc (Commands tab, "Suspend"),"
I wrote: "It's not a processor issue since if I suspend all work, and no processor is used, movies still take some time to load."
Pay attention.
I'm confirming the 1 minute time interval delay.
Futher more: I believe we are not beginners in computer science. I believe we are working on a computer for more than 3 years and perhaps, on multiple OS's. Hmm, haven't tried BOINC on Linux yet. In that way I expect that people know that if an issue is being presented, is already tested for common causes by that person (he who reports it). In any case, this is what I expect of myself.
____________
ID: 15814 | Rating: 0 | rate:
/
Rhiju Forum moderator Project administrator Project developer Project scientist Joined: Jan 8 06 Posts: 223 ID: 48256 Credit: 3,546 RAC: 0
We're super glad to hear that the memory requirements are lower for 5.12. The rosetta development team has made a real effort to reduce the memory requirements. The memory will even shrink a bit more in apps we're testing over the next week.
We are concerned, however, about the effect on WMP and other things. We have added nothing since 5.12 that would cause the problem... maybe something snuck into the BOINC graphics api. I'm contacting the BOINC team.
Finally, Dimitris and Feet1st, are you only seeing the page faults on that workunit? (JUMP_CLOSE_CHAINBREAK_ALLBARCODE_...) If so, I'll cancel that WU. Also, how are you monitoring page faults?
Two questions:
(1) What level of memory use are you seeing for rosetta@home (with graphics) on your computers?
Under Windows, Rosetta used to take 60MB to 160MB of memory (with some HUGE WUs which needed up to 250MB a month ago) and 300-450MB of virtual, until recently. Latest v5.12 seems to have drastically reduced memory to "just" 60-75MB and 300-320MB virtual.
For monitoring processes under Windows I recommend the excellent free Process Explorer util.
Oddly, under Linux, Rosetta takes about HALF of that (30-40MB and up to 85MB).
This is ofcourse empirical, from casual observation.
ps u -U boinc
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
boinc 2120 0.0 0.5 7468 3756 ? S Apr27 0:17 ./boinc_client
boinc 4763 10.1 11.0 116472 82112 ? SN May09 118:20 rosetta_beta_5.12
boinc 4764 0.0 11.0 116472 82112 ? SN May09 0:00 rosetta_beta_5.12
boinc 4765 0.0 11.0 116472 82112 ? SN May09 0:00 rosetta_beta_5.12
boinc 19053 69.3 4.0 94640 30264 ? RN 13:43 182:12 rosetta_5.12_i686
boinc 19054 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
boinc 19055 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
boinc 19056 0.0 4.0 94640 30264 ? SN 13:43 0:00 rosetta_5.12_i686
(2) Should we disable the graphics in the next release to reduce memory use for larger proteins? (at ~100Mb per work unit, there should be no problems on most machines).
Please don't disable graphics. Remember that for many crunchers, the "reward" is the cool screensaver graphics and the credits.
I don't know if you did some changes in latest v5.12, but as I said memory requirements under Win seem to half of what I used to see for the past 2 months. Maybe it's the algorithm (jumping), rather than an issue of the graphics thread, you should know best.
I just read a post from another fellow in Greece that on his PC too there's a 1min delay in starting to play video when Rosetta 5.12 is running (tried under both BOINC 5.2.13 and 5.3.31) and he tried WMP, media player classic, power dvd.
hello, i am the one Dimitris is refering.
there is a major problem here, i can run nothing that containg ANY multimedia contect especially videos.
suspending the project does not affect the problem.
i have to shut down boinc in order to run my videos.
the system is an A64
3500@2800
2GB ram
2Xraptors oin raid 0
X1900XT
Rhiju, I don't have any 5.12 crunching as yet. Not seeing the problem at the moment, and all of my observations were on prior releases. I was thinking of starting a thread and saying "Let's just talk about those that pull > 1m faults per hour of runtime", because I was seeing maybe 30% of WUs were doing just that... on PCs with 1GB of memory.
I pushed some screenshots of 3 WUs up to the web for you to see. I will try to be more dilligent when the 5.12 WUs come through my cache. Is there more info. I should try to capture? I've got WU name, runtime, total page faults, R@H version... anything else?
[edit] I think the screen shots will answer your question about how (at least for my machines) I've been monitoring page faults. I'm looking at task manager PF Delta and "Page Faults" (the total for the PID).
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
are you only seeing the page faults on that workunit? (JUMP_CLOSE_CHAINBREAK_ALLBARCODE_...) If so, I'll cancel that WU. Also, how are you monitoring page faults?
For WinXPpro I monitor page faults from ProcessExplorer util (mentioned earlier). One can click on the "System Information" button which presents stats about the whole system (paging, CPU, I/O etc). When I saw "Page Fault Delta" (i.e. change in page faults per second) high, I right-clicked on the running processes until I found the one causing it.
I think the screen shots will answer your question about how (at least for my machines) I've been monitoring page faults. I'm looking at task manager PF Delta and "Page Faults" (the total for the PID).
in about 22 miniutes i have 242.000 page fault increasing about 650/s ....
I think the screen shots will answer your question about how (at least for my machines) I've been monitoring page faults. I'm looking at task manager PF Delta and "Page Faults" (the total for the PID).
in about 22 miniutes i have 242.000 page fault increasing about 650/s ....
sorry the process is running for about 7 minutes after the last reboot
____________
A64 3500@2727MHz - A64 3200@2400MHz - P4 3.0 northy@3.6MHz
Sempron 2600+@2100MHz - Celeron 2.8 northy @3.6MHz
P4 2.4@2.8MHz - Celeron 2.66@3.0MHz - AthlonXP 2000+@1866MHz
[redguard:] I believe we are not beginners in computer science. I believe we are working on a computer for more than 3 years and perhaps, on multiple OS's. Hmm, haven't tried BOINC on Linux yet. In that way I expect that people know that if an issue is being presented, is already tested for common causes by that person (he who reports it). In any case, this is what I expect of myself.
Many with 1,400 credits and 2 posts on the message board ARE computer novices... if that. Certainly one can't presume 3 years of experience on the part of someone who posts.
I apologize for missing your comment about having tried suspending things first... now, could you use your extensive computer experience and see if you can understand the main point of my comments, which was the memory utilization?
A suspended task does not immediately give up all of it's memory. In fact, a suspended task, if indeed the BOINC Manager successfully got it to suspend, would ONLY have memory left as a point of contention, since it isn't using any disk IO, nor CPU time. I've seen cases where BOINC Manager does NOT get the WU to suspend. That was why I suggested confirming it has done so via the task manager.
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
ID: 15840 | Rating: 0 | rate:
/
Rhiju Forum moderator Project administrator Project developer Project scientist Joined: Jan 8 06 Posts: 223 ID: 48256 Credit: 3,546 RAC: 0
Hi all:
Looks like the high rate of page faults is a separate issue that's been around for a while, since at least 5.01 according to Feet1st's screen shots. Thanks for bringing this to my attention. Being a Windows novice, I didn't see that the task manager would give a PF column (it wasn't on by default on my test machine!). So I'll now look into that. My current WUs aren't showing a high rate of PF's, though.
But the main issue now is the severe slowdown some users are seeing with 5.12, and didn't see with 5.07. We think this problem is due to new debugging code that the Boinc team helped us add to the Windows build. For those of you who have seen this slowdown in Windows Media Player or Internet connections, could you possibly attach to ralph? I am posting a new app there (5.13) which is the same as 5.12 but without the new debugging stuff. Hopefully it has the fix. But I need your input! I'm sending out some jobs now with 5.12, and in about half an hour I'll send more out with 5.13.
Rhiju, I don't have any 5.12 crunching as yet. Not seeing the problem at the moment, and all of my observations were on prior releases. I was thinking of starting a thread and saying "Let's just talk about those that pull > 1m faults per hour of runtime", because I was seeing maybe 30% of WUs were doing just that... on PCs with 1GB of memory.
I pushed some screenshots of 3 WUs up to the web for you to see. I will try to be more dilligent when the 5.12 WUs come through my cache. Is there more info. I should try to capture? I've got WU name, runtime, total page faults, R@H version... anything else?
[edit] I think the screen shots will answer your question about how (at least for my machines) I've been monitoring page faults. I'm looking at task manager PF Delta and "Page Faults" (the total for the PID).
____________
ID: 15844 | Rating: 0 | rate:
/
Rhiju Forum moderator Project administrator Project developer Project scientist Joined: Jan 8 06 Posts: 223 ID: 48256 Credit: 3,546 RAC: 0
Quick note: We need to do some maintenance on ralph, so 5.13 won't be ready there in the next half hour -- but hopefully later this evening.
Hi all:
Looks like the high rate of page faults is a separate issue that's been around for a while, since at least 5.01 according to Feet1st's screen shots. Thanks for bringing this to my attention. Being a Windows novice, I didn't see that the task manager would give a PF column (it wasn't on by default on my test machine!). So I'll now look into that. My current WUs aren't showing a high rate of PF's, though.
But the main issue now is the severe slowdown some users are seeing with 5.12, and didn't see with 5.07. We think this problem is due to new debugging code that the Boinc team helped us add to the Windows build. For those of you who have seen this slowdown in Windows Media Player or Internet connections, could you possibly attach to ralph? I am posting a new app there (5.13) which is the same as 5.12 but without the new debugging stuff. Hopefully it has the fix. But I need your input! I'm sending out some jobs now with 5.12, and in about half an hour I'll send more out with 5.13.
Rhiju, I don't have any 5.12 crunching as yet. Not seeing the problem at the moment, and all of my observations were on prior releases. I was thinking of starting a thread and saying "Let's just talk about those that pull > 1m faults per hour of runtime", because I was seeing maybe 30% of WUs were doing just that... on PCs with 1GB of memory.
I pushed some screenshots of 3 WUs up to the web for you to see. I will try to be more dilligent when the 5.12 WUs come through my cache. Is there more info. I should try to capture? I've got WU name, runtime, total page faults, R@H version... anything else?
[edit] I think the screen shots will answer your question about how (at least for my machines) I've been monitoring page faults. I'm looking at task manager PF Delta and "Page Faults" (the total for the PID).
In the Task Manager, click "View", then click "Select columns", add those of interest.
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
MSI K8t Neo motherboard, Athlon 64 754 pin 3000+ cpu, multiple sata drives; 1 Gig ram, MSI fx5600 video card (nvidia drivers). My system has been less responsive than normal for a few days; and actually takes 15 seconds after right clicking on an .avi file before I get the windows options. Second time: 6 seconds.
After totally killing off Rosetta and Boinc: Less than 1 second.
WinXP sp2, and all the updates except the last Windows Genuine Advantage Notification installed. Jump_allbarcodes_antiparallel_1tul_save_all_out_491_15535_0 was only eating up about 49Megs..
If still needed tonight when I return home, I'll connect up to Ralph and see if it acts the same.
____________
ID: 15850 | Rating: 0 | rate:
/
Rhiju Forum moderator Project administrator Project developer Project scientist Joined: Jan 8 06 Posts: 223 ID: 48256 Credit: 3,546 RAC: 0
Perfect. We'd be very grateful if you attached to ralph. I've queued up a lot of work. Ralph 5.13 is up!
MSI K8t Neo motherboard, Athlon 64 754 pin 3000+ cpu, multiple sata drives; 1 Gig ram, MSI fx5600 video card (nvidia drivers). My system has been less responsive than normal for a few days; and actually takes 15 seconds after right clicking on an .avi file before I get the windows options. Second time: 6 seconds.
After totally killing off Rosetta and Boinc: Less than 1 second.
WinXP sp2, and all the updates except the last Windows Genuine Advantage Notification installed. Jump_allbarcodes_antiparallel_1tul_save_all_out_491_15535_0 was only eating up about 49Megs..
If still needed tonight when I return home, I'll connect up to Ralph and see if it acts the same.
There seems to be more discussion here than on ralph boards so I'll add my bit.
Dell mb P4 3.2, HT but using only 1/2 of it for boinc, 1GB ram, ati radeon 9800
Started noticing degradation around the time of ralph 5.11 but didn't pin it to ralph until 5.12.
I do look at ralph graphics occasionally and use the boinc screensaver limited to 5 minutes, but when I was troubleshooting this I wasn't looking at graphics and was too active for the screensaver to kick in. Would hate to see the graphics go, unless they're made optional.
Currently task-damager says ralph's memory usage is about 72MB with VM size 118MB. And yes, it's showing over 2K-4K+ (!) PF every second, which is not wonderful and I'd really like to know what it's doing. This wu has been running 40 cpu minutes and has 2.7 million page faults. Thankfully the memory size is growing only slowly, or this box would be on its lips by now.
I set my prefs down to 2 hours to get rid of this last 5.12 wu then will restart boinc and start some ralph 5.13 if there are any left by then, to see if things improve. Any further comments from me will go to the ralph boards.
____________
ID: 15855 | Rating: 0 | rate:
/
Moderator9 Forum moderator Project administrator Joined: Jan 22 06 Posts: 1014 ID: 53254 Credit: 0 RAC: 0
While I have no doubt of the reports in this thread. I cannot duplicate the problem on a fairly common Dell laptop. I can't get it to show this degradation at all. It will even ingest video through the firewire port and convert it to MPG2 real time without dropping frames. When it is doing that I can see BOINC yielding the CPU, and coming back in when the video turns off.
Checking the Page faults, it is generating some but I would expect a few as the process runs. The application is using around 107MB of memory. But this is a really clean system without a lot of software running other than windows, and BOINC.
I am not seeing any of this problem on a powerBook, or a G4 dual either.
____________
Moderator9
ROSETTA@home FAQ
Moderator Contact
ID: 15857 | Rating: 0 | rate:
/
Thomas F. Bates IV Joined: May 10 06 Posts: 5 ID: 81910 Credit: 448,002 RAC: 290
Just downloaded and installed rosetta 5.12 tonight. Experienced a problem with unresponsive apps after letting PC sit idle for a while, but killing BOINC fixed it.
P4 2.7GHz 1GB RAM, Win2K Pro SP4, ATI Radeon 9600. Not using BOINC screensaver, just a blank screen w/ PW. However I did take a peek at the cool graphics for a bit.
Just downloaded and installed rosetta 5.12 tonight. Experienced a problem with unresponsive apps after letting PC sit idle for a while, but killing BOINC fixed it.
P4 2.7GHz 1GB RAM, Win2K Pro SP4, ATI Radeon 9600. Not using BOINC screensaver, just a blank screen w/ PW. However I did take a peek at the cool graphics for a bit.
Tom, you're new! Welcome! Sounds like there was a snag in this v5.12 which was just released. They are working on the problem and hope to have a fix (a new version) available in the next day or two.
In the meantime, you said ending BOINC fixed it, are you running any other BOINC projects? Or just Rosetta? I mean do you have experience using BOINC and other projects?
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
ID: 15871 | Rating: 0 | rate:
/
Thomas F. Bates IV Joined: May 10 06 Posts: 5 ID: 81910 Credit: 448,002 RAC: 290
I was a longtime SETI@home fan a few years ago. I enjoyed watching my stats approach 99.5%. Then one summer I needed to shut computers off here at home while I was at work, to cut down on air conditioning costs. But I was recently intrigued when someone mentioned BOINC and the possibility of doing something in addition to looking for signs of alien life. :-) So I'm back. With some faster computers. :-)
After playing with this problem further, it appears that Rosetta and Predictor both exhibit this problem, whereas boincsimap, SETI and Einstein do not. I've been running the last two for about two weeks and hadn't noticed any problems. I just signed up for Rosetta, Predictor, and boincsimap tonight and started seeing the problem.
I used the suspend feature to force boinc to switch between projects. If either Rosetta or Predictor are running, I see the problem. boincmgr itself is not affected, so I can get in and suspend the project and see the results right away. Even taskmgr is hard to switch to!
The problem does not seem to occur on my little backup server - an eMachines 2.6GHz Celeron with 256MB RAM and on-board graphics running Win XP. I tried the graphics too.
I can confirm the delay in video playback using media player classic while 5.12 is running. It doesn't appear to slow down any of my other apps. I also found that watching the graphics sometimes causes the video card to stop responding. Eventually things go back to normal with an error message from the driver (Catalyst version 5.13; ATI X850 XT) stating that the video card stopped responding to the driver. 4.7% into my current result, this happened 4 times in a row and by the fourth time, it had reverted to software rendering so I had to restart my machine.
____________
ID: 15878 | Rating: 0 | rate:
/
David Baker Forum moderator Project administrator Project developer Project scientist Joined: Sep 17 05 Posts: 637 ID: 122 Credit: 214,854 RAC: 0
Rhiju already posted information on this, but we (ie, Rom) know what the problem is and I think Rhiju has already compiled with the previous Boinc app so the problem will gone soon. Rom explained to me on the phone why his error trapping addition to the Boinc API caused problems with certain not highly commercialized applications, but I didn't understand well enough to explain it! (I should point he is
adding this improved error trapping to help track down our few remaining problems).
It is probably the code I added for trapping messages to the debugger
viewport.
I'll make trapping debugger messages an optional thing that has to be turned
on in the registry.
----- Rom
-----Original Message-----
From: Rhiju Das
Sent: Wednesday, May 10, 2006 12:40 PM
To: Rom Walton
Cc: David Baker; David Kim
Subject: slowdowns?
Hi Rom:
We've been getting a lot of user complaints about the newest
rosetta apps causing severe slowdowns on people's computers in going
from 5.07 to 5.12. Some links below, and sample messages.
We haven't added any threads or objects from 5.07 to 5.12. I'm
trying to figure out whether our current work units are causing the
problem. But I also suspect that it could be something in the new
boinc api... can you think of any changes that might slow down a PC?
So that's the problem huh. No wonder my Windows Media Classic + ffdshow stopped when trying to open a video file, after the application is upgraded to 5.12. I nearly formatted the computer because of that.
I think this affect all DirectShow-based players because VLC has no problems with it.
____________
Campeones everywhere!
For history :
this morning the problem appeared on 2 other pcs, an p4 3.0@3.6 1GB ram and a A643200@2400MHz with 512MB ram.
my other 2 pcs are not affected yet.
It is NOT limited to video applications, because I don't use them, and I could repeatably hose my pc after a clean reboot with ralph suspended but other boinc projects running, by running ralph 5.12. Processes as vanilla as notepad took 1 minute to start (and the pc was not memory-starved), and at least one non-media app was badly broken. None of this is happening so far with ralph 5.13, so *big* thanks to Rom.
I hope he will put some details in his blog, because that vendor across the lake from the Rosetta lab will need spoon-feeding to make their app(s) less fragile.
The 5.12 issue also seemed to slow down my printing large documents from Word.
I am processing the remaining 5.12 units in one hour preference to clear them out, and then I will switch back to my normal 12 hour preference for the 5.13 units.
The 5.12 issue also seemed to slow down my printing large documents from Word.
I am processing the remaining 5.12 units in one hour preference to clear them out, and then I will switch back to my normal 12 hour preference for the 5.13 units.
Thank you for fixing this problem so quickly! :)
I do a lot of video editing and my linear editing programs were moving at a snails pace compared to normal while Rosetta/BOINC were running. I had to detach from Rosetta or shut down BOINC completely to get my normal cpu power back.
Interesting. I hadn't the slightest clue that Rosetta was causing the issues I was experiencing... severe WMC + ffdshow lag, general system slowness, etc.
BTW, I run quite a few science apps in BOINC, so nailing down a system issue to a particular one is rather tough.
____________
I do a lot of video editing and my linear editing programs were moving at a snails pace compared to normal
Was that with the 5.12 application? Or the 5.13 that was just released to correct the new memory contention problems?
____________
If having a DC project with BOINC is of interest to you, with volunteer or cloud computing resources, but have no time for the BOINC learning curve,
use a hosting service that understands BOINC projects: http://DeepSci.com
2. HOMO_7486_h003_1_LOOPRLX_IGNORE_THE_REST_7486h003_dec297_1.pdb_497_7_0 WSS: 195 megs, Vmem size: 402.6 megs, PF's 1.8 million and growing at a scary rate of knots.
Hope this helps.
-- Edit --
Needless to say, I'll be watching my systems like a hawk as the 5.13 WU come on line. Since this is not a drop dead showstopper for me, I'll let the 5.12's clear from the cache in their own good time.
____________
ID: 15967 | Rating: 0 | rate:
/
Bob Guy Joined: Oct 7 05 Posts: 39 ID: 3119 Credit: 24,895 RAC: 0
This might be what is going on with page faults. The ordinary hard page fault is the one you get when the page needed is in virtual memory - that's the common understanding of what a page fault is. There is another kind - it is what is probably happening in rosetta - see the following:
Soft faults are relatively inexpensive. These occur when a process tries to access the virtual address, and the operating system can satisfy the request without reading in the page from disk. This can happen with pages that the program wants to be zero (called ‘demand zero’ pages), when a page is written to for the first time (‘copy on write’ pages) or if the page is already in memory somewhere else. The last situation occurs if a file is memory mapped into multiple processes or into multiple locations of the same process, and one of the other file references has already caused the data to be in physical memory.
____________________________
I also believe that the slowdowns have their origin in a bug in the MS Directx SDK. I've experienced similar effects as long as 2 years ago in other applications not related to Boinc whose only common component was the DirectX SDK. I do not have much experience with the DirectX SDK and have never been able to find out if that was actually the problem - it's just a hunch.
ID: 16138 | Rating: 0 | rate:
/
Rhiju Forum moderator Project administrator Project developer Project scientist Joined: Jan 8 06 Posts: 223 ID: 48256 Credit: 3,546 RAC: 0
Hi all: thanks very much for the info on page faults. I can see the high PF rate on my Windows system too especially when I turn on/off graphics a few times. We'll look into whether these are soft or hard page faults... and also whether we can free virtual mem completely when graphics are off.
This might be what is going on with page faults. The ordinary hard page fault is the one you get when the page needed is in virtual memory - that's the common understanding of what a page fault is. There is another kind - it is what is probably happening in rosetta - see the following:
Soft faults are relatively inexpensive. These occur when a process tries to access the virtual address, and the operating system can satisfy the request without reading in the page from disk. This can happen with pages that the program wants to be zero (called �demand zero� pages), when a page is written to for the first time (�copy on write� pages) or if the page is already in memory somewhere else. The last situation occurs if a file is memory mapped into multiple processes or into multiple locations of the same process, and one of the other file references has already caused the data to be in physical memory.
____________________________
I also believe that the slowdowns have their origin in a bug in the MS Directx SDK. I've experienced similar effects as long as 2 years ago in other applications not related to Boinc whose only common component was the DirectX SDK. I do not have much experience with the DirectX SDK and have never been able to find out if that was actually the problem - it's just a hunch.
____________
ID: 16199 | Rating: 0 | rate:
/
william Joined: May 21 06 Posts: 1 ID: 83805 Credit: 3,023 RAC: 0
I HAVE TRIED TWICE NOW AND BOTH FAILED. SO I AM OUT OF THIS PROJECT SORRY GUYS
____________
