Large proteins

Message boards : Number crunching : Large proteins

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Tomcat雄猫

Send message
Joined: 20 Dec 14
Posts: 111
Credit: 1,028,156
RAC: 10,492
Message 94816 - Posted: 19 Apr 2020, 2:15:21 UTC - in response to Message 94811.  

Two questions:
How much RAM will these tasks need?
Will these tasks be run on Android devices?

Thank you.
ID: 94816 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Admin
Project administrator

Send message
Joined: 1 Jul 05
Posts: 5143
Credit: 0
RAC: 0
Message 94824 - Posted: 19 Apr 2020, 4:52:23 UTC - in response to Message 94816.  
Last modified: 19 Apr 2020, 5:07:07 UTC

These larger than usual jobs come from our Robetta structure prediction server. They may or may not be related to COVID-19. The spike protein which was modeled in late January and early February, and continues to be modeled (variants and such), is over 1000 residues, and modeled as a symmetric trimer. The structure has since been determined by Cryo-EM (6vsb, 6vxx, and 6vyb), the latter two from Dr. Veesler's lab at the University of Washington. If the sequence is larger than 1000 residues (2000 is the maximum for a single chain), then the memory bound is set to 4gigs so if your device has enough memory per cpu set to run, then it can run such jobs.
ID: 94824 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Grant (SSSF)

Send message
Joined: 28 Mar 20
Posts: 492
Credit: 1,382,821
RAC: 22,358
Message 94827 - Posted: 19 Apr 2020, 6:00:06 UTC - in response to Message 94824.  

If the sequence is larger than 1000 residues (2000 is the maximum for a single chain), then the memory bound is set to 4gigs so if your device has enough memory per cpu set to run, then it can run such jobs.
Many of the present systems have issues when only a couple of Tasks require 1.3GB of RAM.
Will these larger RAM requirement Tasks be released as a batch, or (ideally), an intermittent stream of one at a time over many hours (depending how many there are of course). Reducing the chances of a system getting more than one to process at a time (i'm thinking a 4GB RAM requirement will make even just 1 of them unsuitable for any Pi system).
Grant
Darwin NT
ID: 94827 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
lakotamm

Send message
Joined: 28 Jun 19
Posts: 22
Credit: 148,806
RAC: 2,068
Message 94850 - Posted: 19 Apr 2020, 10:09:03 UTC

It would be great if there were still units for ARM64 devices with 0,95GB RAM. It seems like the community has found ways to run multiple WUs on these devices in combination with RAM compression. The word is still just getting spread though and I expect the computing power coming from these devices to only increase.

That being said, my laptop with 24GB RAM is ready to take new 4GB WUs with ease.
ID: 94850 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1096
Credit: 4,564,689
RAC: 4,166
Message 94857 - Posted: 19 Apr 2020, 10:50:30 UTC - in response to Message 94824.  

If the sequence is larger than 1000 residues (2000 is the maximum for a single chain), then the memory bound is set to 4gigs so if your device has enough memory per cpu set to run, then it can run such jobs.

Do you plan to insert "long wus" and "short wus" in user's profile (like others project do)??
ID: 94857 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Grant (SSSF)

Send message
Joined: 28 Mar 20
Posts: 492
Credit: 1,382,821
RAC: 22,358
Message 94860 - Posted: 19 Apr 2020, 11:00:23 UTC - in response to Message 94857.  

Do you plan to insert "long wus" and "short wus" in user's profile (like others project do)??
In this case i think "large RAM requirement" and "small RAM requirement" might be the way to go.
Grant
Darwin NT
ID: 94860 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1096
Credit: 4,564,689
RAC: 4,166
Message 94861 - Posted: 19 Apr 2020, 11:05:17 UTC - in response to Message 94860.  

Do you plan to insert "long wus" and "short wus" in user's profile (like others project do)??
In this case i think "large RAM requirement" and "small RAM requirement" might be the way to go.

Yeap. I think it's not important the definition of these wus, but if there will be the possibility to choose.
ID: 94861 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Millenium

Send message
Joined: 20 Sep 05
Posts: 67
Credit: 184,283
RAC: 302
Message 94874 - Posted: 19 Apr 2020, 14:04:57 UTC

Thank you for the news, time to put these 16GB to use!
ID: 94874 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 3994
Credit: 0
RAC: 0
Message 94880 - Posted: 19 Apr 2020, 14:47:17 UTC
Last modified: 19 Apr 2020, 14:48:58 UTC

There will still be a mix of other WUs in addition to these large-protein WUs. If I understand the definition of WU memory bound, it means the 4GB tasks will not be sent to any machine where BOINC is configured to use less than 4GB of memory. Just like the existing WUs, they won't require their maximum memory very often. Crossing the "bound" during the run would cause the BOINC Manager to abort the task. So it is a number you shouldn't actually see happen often.

But I wanted to get people thinking about how long-running, high-memory models will impact the work cache and the BOINC Manager's ability to plan work. To plan your WU runtime preference. Maybe even plan some new resource share to low-memory BOINC projects so the memory requirement of the typical mix of work is reduced.
Rosetta Moderator: Mod.Sense
ID: 94880 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Charles Dennett
Avatar

Send message
Joined: 27 Sep 05
Posts: 102
Credit: 2,070,826
RAC: 2,143
Message 94898 - Posted: 19 Apr 2020, 15:59:21 UTC

Will there be a way to identify these long running tasks based on their name?
-Charlie
ID: 94898 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jim1348

Send message
Joined: 19 Jan 06
Posts: 424
Credit: 19,077,304
RAC: 83,879
Message 94900 - Posted: 19 Apr 2020, 16:05:35 UTC - in response to Message 94880.  

Crossing the "bound" during the run would cause the BOINC Manager to abort the task. So it is a number you shouldn't actually see happen often.

What I have seen in the past on my 12-core Ryzens with 16 GB of memory (11 cores on BOINC) is that if there is not enough memory, then the last WU just won't start up, and I then have only 10 running.
If that is all that happens, no problem. If something crashes, then I think we need to make other arrangements.
ID: 94900 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 3994
Credit: 0
RAC: 0
Message 94902 - Posted: 19 Apr 2020, 16:12:20 UTC - in response to Message 94900.  

Yes the BOINC Manager will suspend one or more tasks as "waiting for memory" as necessary. Eventually the high memory task will be closer to the deadline than other tasks and BOINC will hold enough other threads to allow it to run. But it is based on actual memory requests during the run, not the 4GB memory bound defined for the WU.
Rosetta Moderator: Mod.Sense
ID: 94902 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile WBT112

Send message
Joined: 11 Dec 05
Posts: 11
Credit: 1,117,091
RAC: 1,195
Message 94903 - Posted: 19 Apr 2020, 16:39:01 UTC

Some work preferences would still be nice, assuming that a lot of people just want to crunch without looking into the preferences every few days.
Maybe just send maximum 1-2 of these "monsters" to each user or let them decide manually if they want more.

I don't want my CPU cores idle "waiting for memory" and I can't check all day long.
ID: 94903 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Admin
Project administrator

Send message
Joined: 1 Jul 05
Posts: 5143
Credit: 0
RAC: 0
Message 94908 - Posted: 19 Apr 2020, 17:05:30 UTC

I want to make sure people understand that the long running jobs may be a separate issue and not related. The 12v1n_ tasks are running into rare issues. We are looking into this. There were 2 issues. 1 was large proteins reaching the watchdog time limit before producing a model, and the other issue was with these cyclic peptide 12v1n_ jobs. The former has been dealt with hopefully by extending the watchdog time the latter is likely a protocol issue where for rare cases the modeling trajectory is continuing for a long time due to filtering criteria, but we need to look into this further.
ID: 94908 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1096
Credit: 4,564,689
RAC: 4,166
Message 94912 - Posted: 19 Apr 2020, 17:34:18 UTC - in response to Message 94903.  

I don't want my CPU cores idle "waiting for memory" and I can't check all day long.

+1
ID: 94912 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile [VENETO] boboviz

Send message
Joined: 1 Dec 05
Posts: 1096
Credit: 4,564,689
RAC: 4,166
Message 94915 - Posted: 19 Apr 2020, 17:38:44 UTC - in response to Message 94880.  

There will still be a mix of other WUs in addition to these large-protein WUs. If I understand the definition of WU memory bound, it means the 4GB tasks will not be sent to any machine where BOINC is configured to use less than 4GB of memory.

In which way the server scheduler can recognize the quantity of ram in a host?
For example i have a pc with 24 core and 24 gb of ram. Up to now, not big problems with R@H.
But if you release the "monster ram" wus, i will download only 6 wus and my machine will be full.
Please, give us the possibility to decide.
ID: 94915 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Admin
Project administrator

Send message
Joined: 1 Jul 05
Posts: 5143
Credit: 0
RAC: 0
Message 94916 - Posted: 19 Apr 2020, 17:41:15 UTC - in response to Message 94915.  

We may consider boosting credits for these jobs.
ID: 94916 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 3994
Credit: 0
RAC: 0
Message 94918 - Posted: 19 Apr 2020, 17:46:19 UTC - in response to Message 94915.  

Keep in mind that many of the WUs now are tagged as 2GB, and you are seeing them generally run in less than 1GB. Also keep in mind you are unlikely to pull 4 of them at the same time. I have started a thread to discuss fair credit for completing a monster.
Rosetta Moderator: Mod.Sense
ID: 94918 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Jim1348

Send message
Joined: 19 Jan 06
Posts: 424
Credit: 19,077,304
RAC: 83,879
Message 94924 - Posted: 19 Apr 2020, 18:28:10 UTC - in response to Message 94903.  
Last modified: 19 Apr 2020, 18:36:13 UTC

I don't want my CPU cores idle "waiting for memory" and I can't check all day long.

With virtual cores, it is not much of a loss if you have one out of twelve idle, for example. You aren't losing a full core, only an instruction stream, and the hardware resources will still be in use.
If you are losing two virtual cores, then that is equivalent of a full core. One defense (the best) is to get more memory.

But if they need to get the science done, then the loss of cores for less important work is really their decision, and not a problem for me.

EDIT: As was said, another way is to run a second project that requires less memory. I often use TN-Grid, a gene-expansion project that has some secondary implications for COVID-19 (among a lot of others). It requires only about 56 MB per WU. Just set it to maybe 10% resource share or less, and it will fill in automatically if you have a free core.
ID: 94924 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Sid Celery

Send message
Joined: 11 Feb 08
Posts: 1262
Credit: 23,901,708
RAC: 6,765
Message 94936 - Posted: 19 Apr 2020, 20:17:05 UTC - in response to Message 94918.  

Keep in mind that many of the WUs now are tagged as 2GB, and you are seeing them generally run in less than 1GB. Also keep in mind you are unlikely to pull 4 of them at the same time. I have started a thread to discuss fair credit for completing a monster.

I probably haven't reached the thread announcing that something big has changed, but my laptop (4 cores - 8Gb RAM) was running 2 of the 121vN tasks simultaneously with another queued up to run next and I just spotted that the server had cancelled all three after an update.

I'll catch up shortly
ID: 94936 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
1 · 2 · 3 · Next

Message boards : Number crunching : Large proteins



©2020 University of Washington
https://www.bakerlab.org