When I clear all, the Matlab memory usage drops to 0.5GB. I then load a *.mat file in which the memory requirement is dominated by an object requiring 161MB (from whos). The Matlab memory usage jumps to 1.3GB. I then run a script that processes the data, creating small data objects/structures in the process. From whos, nothing rivals the 161MB object in terms of memory footprint. However, the Matlab's memory footprint creeps up to 2.2GB during processing, then settles down to 1.6GB when done. whos still reveals the overwhelmingly dominant memory user to be the loaded object.
Why does Matlab use so much memory than the data that it is processing? It's about 1000 times more. Is this just to give it the space for intermediate results?
I'm using Windows 7, 64-bit. My code is pretty simple post-processing script to tally up some of the loaded data. It invokes no user-defined functions or 3rd party tools. I understand that readers can't analyze my code to track down the specific causes, but is 1000x memory footprint typical? What are typical reasons for this?
Related
I was reading this article Why mmap is faster than system calls, where the main difference appeared to be mmap's ability to use vector instructions like AVX-2, something system calls can't.
I understand that the SIMD instructions used by GPUs tend to be much wider. A Nvidia warp of size 32 operating on float32 = 1024 bits (?) vs 256 bits of AVX-2. So potentially a 4x speedup. I guess this is not used in traditional discrete gpu settings as host-to-device (and back) copy would outweigh any benefit from wide registers.
However in APUs, GPU shares memory with CPU, eliminating the need for these expensive copies. I was wondering if those GPU instructions can therefore be used to accelerate mmap like vector operations further (numpy is another example). Has it already been done (in M1 mac or any CPUs with integrated graphics)? or can you please detail the architectural issues that prevent this?
You're kind of asking 2 separate questions: whether an OS (or user-space standard libraries?) can use GPGPU to speed up reading from the pagecache (into user-space memory with a read system call, or from an mmaped region). And separately whether GPGPU on normally-allocated process memory (and/or the pagecache) can avoid a copy to memory dedicated to the GPU.
For the 2nd part Apple has said the answer is yes for MacOS on M1 thanks to making the integrated GPU's memory accesses cache-coherent with the CPU. I think AMD made similar suggestions that copying could be avoided in graphics or GPGPU drivers on their APUs (Fusion IIRC?), but IDK if software ever took full advantage.
For the first part; doubtful. Large memory copies are bottlenecked by DRAM bandwidth, not CPU-core <-> L1d cache bandwidth (which scales with SIMD register width). On x86, an AVX2 loop on a single core can come pretty close to maxing out the DRAM bandwidth of an Intel "client" chip (quad-core or similar, not a big xeon with a higher-latency interconnect). Single-core bandwidth (to L3 or DRAM) tends to be limited by the number of outstanding cache misses that a core can track, not by doing the copy with fewer instructions. That mostly helps in terms of seeing farther with the same size out-of-order execution window, to start page walks sooner across page boundaries and stuff like that. See Why is std::fill(0) slower than std::fill(1)? for SSE (16-byte) vs. AVX (32-byte) vectors.
GPU offload would thus not help for large copies. It could only possibly help for small copies, and then it would not leave the copy result hot in L1d cache of the CPU. And/or not be able to take advantage of the source or destination already being hot in L1d cache of a CPU working with the data.
Also, setup overhead (to communicate with the GPU, going outside the current core) would dominate any faster copying for small copies.
In the debugging mode I stop at some breakpoint and do some matrix manipulation in order to test the program. These manipulations are computationally expensive so MATLAB uses the swap space on my linux system. Then, after continuing the program running, the swap space is almost full so MATLAB crushes. Is there a way I could clean the swap at the debugging node? Doing clear all and clear classes makes effect only on RAM memory, but do not affect the swap.
You can't. Swap isn't special, so just work through this as as a regular out-of-memory issue. If you free up memory, you'll indirectly free up the swap that's being used to back it (or avoid having to use swap to supplement it).
Swap space is just an OS-managed backing store for virtual memory. From a normal program's point of view, swap is RAM (just slow RAM) and you don't manage it separately. (Well... you can "wire" pages to prevent them from being swapped out and so on, or use OS APIs to directly manipulate swap, but those are low-level platform-specific details, (like, below malloc), and not exposed to you as a Matlab M-code programmer, and not what you want to do here.) If your Matlab program runs out of memory, that means it's used up or fragmented its process's virtual memory, not something in particular about your swap space. (Unless there's a low-level bug somewhere.)
When this happens, you may need to look elsewhere in your Matlab program (e.g. in global variables, figure handle properties, or other levels of the function call stack) to find additional data that hasn't been cleared yet, or just restart the Matlab process to fix memory fragmentation (which can happen if your code fills up the memory with lots of small arrays).
Like #siliconwafer suggests, memory, whos, and feature memstats are good tools for debugging this. And if you're stopped inside the debugger, realize you can't actually clear everything until you dbquit out of it.
Doing large matrix operations inside the debugger is not necessarily a recoverable operation: if you've modified arrays held in local variables in the stack frame(s) you're working on, but there are still copies of them held in other variables or frames, Matlab's copy-on-write mechanism needs to hold on to both copies of the arrays, and you might be out of luck for that run of the program if you hit your RAM limits.
If clear all and clear classes after exiting the debugger are not recovering enough memory for you, that smells like either memory fragmentation or a C-level memory leak (like in a MEX file). In either case, you need to restart Matlab to resolve it. Avoid the use of large cellstr arrays or other arrays-of-small-arrays to reduce fragmentation. And take a good hard look at your C code if you're using any custom MEX functions.
Or you just might not have enough memory to do the operations you're doing.
I was testing a stand-alone application we developed in Matlab when I noticed that its memory usage, according to Windows Task Manager, was peaking several times above 16gb. I decided to run Matlab's profiler with profile -memory on on the scripts behind the compiled version to see where the memory peaks were occurring, using the exact same input. However, the highest peak memory it found was 2400860.00 Kb, or about 1/4 as much, for the function that essentially acts as the program's main().
Thus, I was wondering if people have noticed huge memory usage differences between running a compiled Matlab program and running the original scripts in Matlab. I noticed it took a lot longer running in Matlab, but I figured that was due to the profiler keeping track of all of the memory allocations and deallocations, rather than reading and writing to a swap space on disk.
To make a real quick answer to this question. Yes, MATLAB compiled applications run with more overhead than MATLAB scripts.
This is because MATLAB deployed applications open up a version of MATLAB which is stored in the memory called the MCR. The MCR runs with more overhead than MATLAB.
One thing that I have found useful in situations like this is to recompile and see if that helps at all. If it doesn't, you could try to lower the memory usage by running calculations in segments.
This might be helpful for better memory usage: http://www.mathworks.com/help/matlab/matlab_prog/strategies-for-efficient-use-of-memory.html
Source:
http://www.mathworks.com/matlabcentral/newsreader/view_thread/306814
Matlab executable too slow
Comment if you have questions.
I've been tasked with writing a program that does streaming sums of vectors into scattered memory locations, at the absolute max speed possible. The input data is a destination ID and an XYZ float vectors, so something like:
[198, {0.4,0,1}], [775, {0.25,0.8,0}], [12, {0.5,0.5,0.02}]
and I need to sum them into memory like so:
memory[198] += {0.4,0,1}
memory[775] += {0.25,0.8,0}
memory[12] += {0.5,0.5,0.02}
To complicate matters, there will be multiple threads doing this at the same time, reading from different input streams but summing to the same memory. I don't anticipate there being a lot of contention for the same memory locations, but there will be some. The data sets will be pretty large - multiple streams of 10+ GB apiece that we'll be streaming simultaneously from multiple SSDs to get the highest possible read bandwidth. I'm assuming SSE for the math, although it certainly doesn't have to be that way.
The results won't be used for a while, so I don't need to pollute the cache... but I'm summing into memory, not just writing, so I can't use something like MOVNTPS, right? But since the threads won't be stepping on each other that much, how can I do this without a lot of locking overhead? Would you do this with memory fencing?
Thanks for any help. I can assume Nehalem and above, if that makes a difference.
You can use spin locks for synchronized access to array elements (one per ID) and SSE for summing. In C++, depending on the compiler, intrinsic functions may be available, e.g. Streaming SIMD Extensions and InterlockExchange in Visual C++.
Your program's performance will be limited by memory bandwidth. Don't expect significant speed improvement from multithreading unless you have a multi-CPU (not just multi-core) system.
Start one thread per CPU. Statically distribute destination data between these threads. And provide each thread with the same input data. This allows better use of NUMA architecture. And avoids extra memory traffic for thread synchronization.
In case of single-CPU system, use only one thread accessing destination data.
Probably, the only practical use for more cores in CPUs is to load input data with additional threads.
One obvious optimization is to align destination data by 16 bytes (to avoid touching two cache lines while accessing single data element).
You can use SIMD to perform the addition, or allow compiler to automatically vectorize your code, or just leave this operation completely unoptimized - it doesn't matter, it's nothing compared to the memory bandwidth problems.
As for polluting the cache with output data, MOVNTPS cannot help here, but you can use PREFETCHNTA to prefetch output data elements several steps ahead while minimizing cache pollution. Will it improve performance or degrade it, I don't know. It avoids cache trashing, but leaves most of the cache unused.
I am suffering from the out of memory problem in mablab.
Is is possible to build a large memory system for matlab(e.g. 64GB ram)?
If yes, what do I need?
#Itamar gives good advice about how MATLAB requires contiguous memory to store arrays, and about good practices in memory management such as chunking your data. In particular, the technical note on memory management that he links to is a great resource. However much memory your machine has, these are always sensible things to do.
Nevertheless, there are many applications of MATLAB that will never be solved by these tips, as the datasets are just too large; and it is also clearly true that having a machine with much more RAM can address these issues.
(By the way, it's also sometimes the case that it's cheaper to just buy a new machine with more RAM than it is to pay the MATLAB developer to make all the memory optimizations they could - but that's for you to decide).
It's not difficult to access large amounts of memory with MATLAB. If you have a Windows or Linux machine with 64GB (or more) - it will obviously need to be running a 64-bit OS - MATLAB will be able to access it. I've come across plenty of MATLAB users who are doing this. If you know what you're doing you can build your own machine, or nowadays you can just buy a machine that size of the shelf from Dell.
Another option (depending on your application) would be to look into getting a small cluster, and using Parallel Computing Toolbox together with MATLAB Distributed Computing Server.
When you try to allocate an array in Matlab, Matlab must have enough contiguous memory the size of the array, and if not enough contiguous memory is available, you will get out of memory error, no matter how much RAM you have on your computer.
From my experience, the solution will not come from dealing directly with memory-related properties of your hardware, but from writing your code in a way that prevents allocation of too large arrays (cutting data to chunks, etc.). If you can describe your code and the task you try to solve, it might be possible to guide you in that direction.
You can read more here:http://www.mathworks.com/support/tech-notes/1100/1106.html