I just ran into a problem with some work right now that is thrashing my hard drive to death. I'm not sure which of the work units is doing it, but my hard drive light is staying continuously lit while Rosetta is processing. If I snooze the client, the hard drive activity light returns to more normal patterns. Someone let some bad wus out it looks like to me. Everything is responding dog slow while running Rosetta because it's thrashing the crap out of my hard drive. If this continues, I will just back to Seti or some other BOINC project because I'm not going to let Rosetta kill my hard drive and make using my computer be a crappy experience.

I just ran into a problem with some work right now that is thrashing my hard drive to death. I'm not sure which of the work units is doing it, but my hard drive light is staying continuously lit while Rosetta is processing. If I snooze the client, the hard drive activity light returns to more normal patterns. Someone let some bad wus out it looks like to me. Everything is responding dog slow while running Rosetta because it's thrashing the crap out of my hard drive. If this continues, I will just back to Seti or some other BOINC project because I'm not going to let Rosetta kill my hard drive and make using my computer be a crappy experience.

If the unit is not too far along just kill it and move to the next one and see if it does the same thing. You killing a unit, or even a batch of units, isn't going to throw the Project off the cliff. The units will just get sent out to other pc's to crunch, some people are running Boinc on an SSD drive so no thrashing for them.

I just ran into a problem with some work right now that is thrashing my hard drive to death. I'm not sure which of the work units is doing it, but my hard drive light is staying continuously lit while Rosetta is processing. If I snooze the client, the hard drive activity light returns to more normal patterns. Someone let some bad wus out it looks like to me. Everything is responding dog slow while running Rosetta because it's thrashing the crap out of my hard drive. If this continues, I will just back to Seti or some other BOINC project because I'm not going to let Rosetta kill my hard drive and make using my computer be a crappy experience.

If the unit is not too far along just kill it and move to the next one and see if it does the same thing. You killing a unit, or even a batch of units, isn't going to throw the Project off the cliff. The units will just get sent out to other pc's to crunch, some people are running Boinc on an SSD drive so no thrashing for them.

I ended up having to dump every wu on that machine, because I couldn't find which one (or series) was thrashing my drive and tying my system in knots. After dumping them, I tried to get some new work but the server only sent me 8 wus to crunch, which were done by the time I woke up this morning. Since the server won't send me more work for another 15 hours or so, I have that machine crunching some Seti work at present. BTW, I am having no thrashing problem while processing the Seti work, so it's not client or machine related.

BTW, I am having no thrashing problem while processing the Seti work, so it's not client or machine related.

Right, SETI's program only uses a small amount of memory, how would it be possible for it to cause your machine to swap memory?

I see you have 7 machines actively crunching, so I wasn't able to determine which one might be the problem... or perhaps you are seeing similar symptoms on several? The bottom line is that Rosetta requires a significant amount of memory, per active task, to run. When physical memory is not available, the operating system makes up the difference by swapping things in and out to the hard drive. By the way you used the term "thrashing" it sounds like you are familiar with the concepts.

Just based on the RACs for your machines, I can say they tend to have many processors. Each active Rosetta task requires it's own memory to run. Much of the memory used is for data that is being processed, rather than the application program itself. So if you have an 8 CPU system, and only run Rosetta, that tends to mean 8 active Rosetta threads, and upwards of 8GB of memory (or more) being recommended. As the tasks run, their actual memory requirements will vary up and down. There is also a large variance between types of tasks.

BOINC allows you to set up preferences to limit the amount of memory and swap space taken by the active BOINC applications. If these limits are exceeded, BOINC will suspend tasks that are consuming a lot of memory and look for other tasks that may be able to run in a smaller footprint. This may mean you run less than 8 tasks for a period of time, but saves the stress on your hard drive. It also means that attaching each machine to more than one project can help your machine have tasks with a variety of memory requirements to place in to the mix of what is active.

When situations such as you describe come up, I always find myself suggesting addition of WCG for a portion of your processing (you can establish "resource shares" to help BOINC understand where you want your machine time to go). For example, on an 8 CPU machine, running WCG at something like 12% would mean that over time, you'd typically have one CPU running a WCG task (most of their sub-projects have low memory requirements... <100MB or so, but you can select which you like as well). And the difference between an active Rosetta task that might consume >750MB for a period of time, and the task using only 100MB means there is breathing room for the 7 others CPUs to each use almost 100MB more before reaching the same point of thrashing.

---

Rosetta Moderator: Mod.Sense

Thanks for the useful posts, Modsense.

---

Thanks for the useful posts, Modsense.

Just looked in TaskManager and Minirosetta uses 712MByte (D)RAM
per task. By far the most, compaired to Einstein@Home, Malaria Control, SETI@Home
LHC@home, GPUGrid, Docking@home, Milkyway, etc. CPDN, uses a lot of memory, too.

Using a fast 8 GByte SD card as ReadyBoost*, can/will help, as I noticed the high memory use, inmediatly.
*(Only on VISTA and WIN 7).
Kind of new around here ;-)

---

Knights Who Say Ni N!

Just looked in TaskManager and Minirosetta uses 712MByte (D)RAM

Yes, and that is not unexpected. But it is on the high side of "normal" overall for the tasks sent over a longer period of time. Also, the server only sends tasks that are expected to use larger amounts of memory to machines that have larger amounts of memory. So part of why one sees tasks that use a lot of memory is because their machine has the memory available. In many cases, a task like that would not be sent to a machine with less RAM.

I guess I'm just trying to point out that this number varies... significantly. You will see other tasks running along in 300MB of memory. And as a task completes a model, everything is written to the results file and the work areas are cleared. So you will see the max memory used often significantly higher than the current memory used. The max being just BEFORE the end of a model.

For example, I'm just looking at a 4 core machine that's been running all of the tasks for several hours:

task 1 current 485MB, max 614MB
task 2 current 493MB, max 831MB
task 3 current 505MB, max 540MB
task 4 current 470MB, max 505MB

and so often times, one task will happen to be one that can run in 500MB mixed in with one that needs 800MB at some point during a run. But, just randomly, sometimes a machine ends up with a set of more of the 800MB tasks. Even then, one task will happen to have just completed a model and dropped it's memory usage, while another is reaching the end and needing more memory. The worst case is what I was trying to describe earlier in this thread. To talk about all tasks using 1GB at the same time is really a worse-case discussion.

FYI, on Windows task manager, you can establish what columns you would like displayed using the "view" pulldown menu and "select columns". I like to see the current and peak memory along with the page fault delta.

---

Rosetta Moderator: Mod.Sense

Thanks for the useful posts, Modsense.

Just looked in TaskManager and Minirosetta uses 712MByte (D)RAM
per task. By far the most, compaired to Einstein@Home, Malaria Control, SETI@Home
LHC@home, GPUGrid, Docking@home, Milkyway, etc. CPDN, uses a lot of memory, too.

Using a fast 8 GByte SD card as ReadyBoost*, can/will help, as I noticed the high memory use, inmediatly.
*(Only on VISTA and WIN 7).
Kind of new around here ;-)

I think that's a good point - Readyboost isn't a replacement for RAM but benefits reads even if you have loads of RAM (and no SSD). If you don't have loads of RAM then it can really improve system responsiveness, as long as the flash device is reasonably quick on reads.

If you do use Readyboost then a few points to note:
Win7 can use multiple devices for RB
Enable write caching on any readyboost devices to improve the write speeds (but then you should eject them from the system tray menu if you want to unplug them).
For 4GB+ you need to format the device as exfat.
I believe USB3 devices tend to give the best bang for buck for readyboost (even on USB2 sockets USB3 drives tend to have quicker response times etc) - good article on Anandtech.

---

Thanks for the useful posts, Modsense.

Just looked in TaskManager and Minirosetta uses 712MByte (D)RAM
per task. By far the most, compaired to Einstein@Home, Malaria Control, SETI@Home
LHC@home, GPUGrid, Docking@home, Milkyway, etc. CPDN, uses a lot of memory, too.

Using a fast 8 GByte SD card as ReadyBoost*, can/will help, as I noticed the high memory use, inmediatly.
*(Only on VISTA and WIN 7).
Kind of new around here ;-)

I think that's a good point - Readyboost isn't a replacement for RAM but benefits reads even if you have loads of RAM (and no SSD). If you don't have loads of RAM then it can really improve system responsiveness, as long as the flash device is reasonably quick on reads.

If you do use Readyboost then a few points to note:
Win7 can use multiple devices for RB
Enable write caching on any readyboost devices to improve the write speeds (but then you should eject them from the system tray menu if you want to unplug them).
For 4GB+ you need to format the device as exfat.
I believe USB3 devices tend to give the best bang for buck for readyboost (even on USB2 sockets USB3 drives tend to have quicker response times etc) - good article on Anandtech.

Thanks for your reply, I'm using a 8 GByte SD card(fastest-series), formatted as NTFS. Also noticed, another Device can be added.
It now uses 7.8GByte and HDD'(s) are quiet, no SSD, yet.
And it's possible to have multiple browser-pages 'open' whithout hick-ups, also make a screendump open PAINT and store it.
This INTEL DP67BG Mobo, also has USB 3.0 and eSATA. Copying, while BOINC is running, is a show-stopper, though!

---

Knights Who Say Ni N!