`out of memory` error

@Tianjing Zhao

In Julia, when we run the same code repeatedly, the garbage collection works as follows:

Steps	1. open	2. create matrix	3. repeat	4.repeat	...repeat...
Running		x = rand(10000,10000)	x = rand(10000,10000)	x = rand(10000,10000)	...
Memory(MB)	89.2	866.8	1628.4	1628.4	1628.4

Note that the Memory(MB) information is obtained from Task Manager(Win10), which shows total physical memory reserved by Julia. You can also check free memory by running versioninfo(verbose=true) or Sys.free_memory()/2^20.

From the above table, when we create a matrix, it will definitely use memory, and when we repeat it for the first time (step 3), the memory doubled. But for the consecutive repeating (step 4,5...), the memory doesn't change.

So if you repeat your code for the first time, you will double the memory. In this situation, restarting the kernel will solve the problem. Here is a trick to avoid double memory: change to garbage (step iii) and collect the garbage via GC.gc()(step iv) before repeating. See the below example:

Steps	i. open	ii. create matrix	iii. set to zero	iv. collect garbage	v. repeat
Running		x = rand(10000,10000)	x=0	GC.gc()	x = rand(10000,10000)
Memory(MB)	86.9	865.1	865.1	98.6	861.6