Is there a way to request work units that require large memory models? I now have a few systems with more than 1GB RAM per core and would like to see that RAM put to good use. Currently the largest work unit I have is using about 650MB and most are using less than 400MB.

---

Thx!

Paul

Not at this time. But, with the new servers in place, that sort of change should now be more approachable than it was previously.

---

Rosetta Moderator: Mod.Sense

I looked at one of my machines this morning & it had at least 10 processes using more than 1GB of RAM each! I thought the project team just started throwing me larger processes. This is the fist time I have seen my machine using more than 32GB of RAM. I love to see better resource utilization.

---

Thx!

Paul

That was more of a coincidence. However, on occasion, there are tasks that require more memory, and rather than allocating them to a specific subgroup, the Project Team was able to precent them from going out to systems with low memory. With the new servers, they may be able to make the scheduler more proactive in assigning such tasks.

---

Rosetta Moderator: Mod.Sense

I would love to volunteer my machines to have the resource intensive work units =D

Will this be possible on the new servers?

This is my fault. I was asked by someone in the lab to increase the length limit on Robetta, our publicly available structure prediction server, so I did but since realized the memory issue. I've reverted the length limit until we can figure out a good approach to handle them. Sorry. There may still be some jobs left over. These are medically relevant targets.

I caught these tasks using over 1GB of memory on Windows 7.
pq170720_Sh29_1_DHR09_000603_0050_C_0001...
D2D2F222xtal_AB_5_3gaf_3s55_0007_0001.pdb_unaD2D2F222_17_07_19_36_34_globalDocking_5_SAVE_ALL_OUT...

---

Rosetta Moderator: Mod.Sense

It's a problem to introduce, in user profile's, a button with something like "Permit wus with high usage memory"??
Every user can decide or not to accept these wus.

---

On the list of things to do. great idea. I'll remind/alert the lab about the memory issues. thanks.

There is no need to introduce a button for tasks with high memory usage. BOINC server already handles this out of the box. The server sends only such tasks to a host that have rsc_memory_bound less than the memory limit set in host's/user's computing preferences. If user doesn't want memory filled with tasks all he or she needs to do is lower the memory limit in computing preferences.

That does mean that the project needs to fill rsc_memory_bound accurately. The few tasks that I sampled had rsc_memory_bound=1e8 bytes but actual memory usage was 300-400 MB. Naughty.

On a related topic: I've found that current WUs use considerably less RAM than they did before. If I'm not wrong... R@H WUs used to take about half a GB per WU (thus the popular "1GB-per-core rule of thumb").

---

The amount of memory used by each WU type varies widely as it depends on the simulation programmed by the scientific that created it.

Now, I'm seeing up to 1.3 Gb in many "yh170802_" units, close to 1GB in "JWdesRMf_closed_" units and around 200Mb in "A211_" ones

Now, I'm seeing up to 1.3 Gb in many "yh170802_" units,

Interesting. I never bothered to look before. But BoincTasks shows a "yh17802" using 1329 MB of real memory, and 1508 MB of virtual memory.
I suppose the "real memory" is the one to pay attention to for DRAM purposes. Fortunately, with 32 GB of main memory, it has not been an issue.

---

New king :)

"JWC6-62dock-new..." up to 1,5Gb RAM occupation

I stand corrected... these "yh" WUs are definitely RAM hoarders.

---

Hi,

I'm the culprit behind the "yh..." jobs!
I apologize if it's causing any issues, I have set rsc_memory_bound to be 2gb, but I will set it higher in the future to make sure I don't top out anyone's computers.

These jobs are larger symmetrical proteins, we are testing a computational method to only "fold" a portion of the larger structure.
Even though we only modify a portion, there are just so many atoms there that it seems to eat up a lot of memory.

Thanks for your precious resources though!
Please do let us/me know if anything we do is causing problems.

As others have mentioned, some of us have enough memory to run 2Gb jobs, and will gladly do it. If it works better for you guys, it might be worth designing a way for us to request such jobs.

I have no problem running high-memory WUs :)

---

There is no need to introduce a button for tasks with high memory usage. BOINC server already handles this out of the box. The server sends only such tasks to a host that have rsc_memory_bound less than the memory limit set in host's/user's computing preferences. If user doesn't want memory filled with tasks all he or she needs to do is lower the memory limit in computing preferences.

I continue to prefer the idea of button in profile's page...

---

There is no need to introduce a button for tasks with high memory usage. BOINC server already handles this out of the box. The server sends only such tasks to a host that have rsc_memory_bound less than the memory limit set in host's/user's computing preferences. If user doesn't want memory filled with tasks all he or she needs to do is lower the memory limit in computing preferences.

I continue to prefer the idea of button in profile's page...

I think the profile page is a much better place for it too. Lowering the memory limit in computing preferences is a very clumsy way to do it, and most people would never think of it. Furthermore, it would apply to all projects you run, not just Rosetta.

I am all for using more memory too if some group of tasks can be run more efficiently that way. 3 GB would work for me. And I could do more if necessary by using an app_config.xml to limit the number run concurrently.

---