Why is MATLAB R2019b very slow when clearing large variables from memory?

I have a workstation with Intel i7 9900K and 64 GB (DDR4, 3200 MHz) of RAM. My recent project requires me to load 3-8 large structure arrays into the memory, occupying 10-40 GB of memory in total. I find that MATLAB (R2019b update 3) gets very slow when clearing these large variables from workspace. The "clear" command takes 30-60 seconds to finish, and from the Windows task manager I can see the memory usage gradually goes down. This happens even when aborting a function which loads the variables inside. The system has a pagefile of around 9 GB on an SSD but the memory usage (including all programs) never goes beyond 85% of total memory. I hope to know whether such slow clearing speed is a expected MATLAB behavior?

[Andrew Janke]

Can you tell us more about the format of these structure arrays? E.g. do they have lots of sub-structures or cellstr arrays in them? In Matlab, clearing a single numeric array of any size is a very fast operation. But clearing arrays with lots of sub-arrays can be slow. And in structs and cells, each field contents or cell contents is a separate array (mxArray). The time that it takes to clear this data is not proportional to the total memory size in GB, but rather proportional to the total number of mxArrays in your data structure.

[yunhui zhou]

Yes, one large structure array contains many substructure arrays (for example, three 50000*1 struct), and a single substructure contains 19-20 fields of numeric arrays and cell arrays.

[Guillaume]

That's 50000*20 arrays that all need to be freed individually. That's a lot of arrays!

Also, if you're not on 2019b update 4, see if installing the update improves performance.

[Walter Roberson]

Symbolic expressions stored in the symbolic engine can also be pretty slow to be removed, even when you quit MATLAB.