Search
Thursday, 11th of October 2018, 15:17:39 UTC
15:24:25
drmeister
::notify kpoeck_ Thank you - I look forward to the clx patches for clasp
15:24:26
Colleen
drmeister: Got it. I'll let kpoeck_ know as soon as possible.
15:49:34
drmeister
Bike, Shiho - how are things going?
15:52:04
Bike
i thought of something that could be the problem but it's looking like it wasn't
15:52:34
shiho
I tried to compare nonbond parameters in Alex's computer, but it looks we need more memory...
15:55:50
drmeister
shiho: How do you know it needs more memory?
15:57:03
drmeister
That is something else.
15:57:37
shiho
I tried other topology files (only protein). It works.
15:58:08
drmeister
Clasp can crash for other reasons - other than memory. On macOS I've rarely seen it crash because of memory.
15:58:29
drmeister
Bike: Can you show Shiho how to log in from a terminal and run 'top'? Then you can monitor memory.
16:00:15
Bike
i can just watch in process monitor on the screen
16:01:23
drmeister
Ok, but then you have to watch on your computer what shiho is doing from her computer.
16:01:30
drmeister
That works but it ties you both up.
16:02:41
Bike
shiho is running from emacs so top isn't even interactive, let alone colored
16:03:04
jackdaniel
C-u slime ansi-term for the rescue
16:05:56
jackdaniel
I need some rest, too many silly mistakes \o
16:08:18
Bike
shiho's run got up to ~37GB memory usage and then died
16:08:33
Bike
"Inappropriate ioctl for device"
16:08:42
Bike
may or may not be the error
16:08:54
Bike
but i'd say if you're using that memory you probably need to change the program
16:09:52
drmeister
This is a one-time or rare exercise in comparing two very large datasets to each other.
16:10:16
drmeister
Shiho: Can you tell me how many atoms are in the protein?
16:10:23
drmeister
(chem:number-of-atoms *agg*)
16:11:02
drmeister
Then square that and look at the energyNonbond.h EnergyNonbond C++ structure, estimate its size and multiply that size by the number-of-atoms^2
16:11:09
drmeister
Does that get us anywhere near 37GB?
16:11:46
shiho
Number of Nonbond 42122356
16:12:03
drmeister
https://www.irccloud.com/pastebin/6TfIEKjz/
16:12:57
drmeister
So... 96 bytes per EnergyNonbond
16:13:27
drmeister
So 42122356*100 = 4,212,235,600
16:13:48
drmeister
Something is wrong if we are using 37GB of memory.
16:14:39
drmeister
Measure the memory just before you run the function that unpacks the nonbond excluded atom list.
16:14:49
drmeister
Then start the function and see if the memory blows up.
16:15:27
drmeister
Then add print statements to print how many NonbondEnergy terms you are creating and print once the sizeof(NonbondEnergy) and do some arithmetic to figure out what is going on.
16:15:50
drmeister
If we had MPS working then I could use (ROOM) to walk memory and tell you how much memory each data type is taking. #$%*#%*
18:39:16
heisig
Bike: By the way, my load time value hoisting code is now in Cleavir. Just in case you need it someday.
18:39:36
Bike
noted. maybe i'll take a closer looka t it.
18:41:56
heisig
No hurry. I just mentioned it to prevent duplicate work.
2:26:26
drmeister
https://gist.github.com/
2:26:34
drmeister
https://gist.github.com/drmeister/e2b4648eb71b5c7232e8dbb09d525b1d
2:26:52
drmeister
That defines everything that inherits from everything.
2:36:17
beach
Good morning everyone!
Friday, 12th of October 2018, 3:17:39 UTC