I have a c++ program that uses several very large arrays of doubles, and I want to reduce the memory footprint of this particular part of the program. Currently, I'm allocating 100 of them and they can be 100 Mb each.
Now, I do have the advantage, that eventually parts of these arrays become obsolete during later parts of the program's execution, and there is little need to ever have the whole of any one of then in memory at any one time.
My question is this:
Is there any way of telling the OS after I have created the array with new or malloc that a part of it is unnecessary any more ?
I'm coming to the conclusion that the only way to achieve this is going to be to declare an array of pointers, each of which may point to a chunk say 1Mb of the desired array, so that old chunks that are not needed any more can be reused for new bits of the array. This seems to me like writing a custom memory manager which does seem like a bit of a sledgehammer, that's going to create a bit of a performance hit as well
I can't move the data in the array because it is going to cause too many thread contention issues. the arrays may be accessed by any one of a large number of threads at any time, though only one thread ever writes to any given array.
It depends on the operating system. POSIX - including Linux - has the system call madvise to do improve memory performance. From the man page:
The madvise() system call advises the kernel about how to handle paging input/output in the address range beginning at address addr and with size length bytes. It allows an application to tell the kernel how it expects to use some mapped or shared memory areas, so that the kernel can choose appropriate read-ahead and caching techniques. This call does not influence the semantics of the application (except in the case of MADV_DONTNEED), but may influence its performance. The kernel is free to ignore the advice.
See the man page of madvise for more information.
Edit: Apparently, the above description was not clear enough. So, here are some more details, and some of them are specific to Linux.
You can use mmap to allocate a block of memory (directly from the OS instead of the libc), that is not backed by any file. For large chunks of memory, malloc is doing exactly the same thing. You have to use munmap to release the memory - regardless of the usage of madvise:
void* data = ::mmap(nullptr, size, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
// ...
::munmap(data, size);
If you want to get rid of some parts of this chunk, you can use madvise to tell the kernel to do so:
madvise(static_cast<unsigned char*>(data) + 7 * page_size,
3 * page_size, MADV_DONTNEED);
The address range is still valid, but it is no longer backed - neither by physical RAM nor by storage. If you access the pages later, the kernel will allocate some new pages on the fly and re-initialize them to zero. Be aware, that the dontneed pages are also part of the virtual memory size of the process. It might be necessary to make some configuration changes to the virtual memory management, e.g. activating over-commit.
It would be easier to answer if we had more details.
1°) The answer to the question "Is there any way of telling the OS after I have created the array with new or malloc that a part of it is unnecessary any more ?" is "not really". That's the point of C and C++, and any language that let you handle memory manually.
2°) If you're using C++ and not C, you should not be using malloc.
3°) Nor arrays, unless for a very specific reason. Use a std::vector.
4°) Preferably, if you need to change often the content of the array and reduce the memory footprint, use a linked list (std::list), though it'll be more expensive to "access" individually the content of the list (but will be almost as fast if you only iterate through it).
A std::deque with pointers to std::array<double,LARGE_NUMBER> may do the job, but you better make a dedicated container with the deque, so you can remap the indexes and most importantly, define when entries are not used anymore.
The dedicated container can also contain a read/write lock, so it can be used in a thread-safe way.
You could try using lists instead of arrays. Of course list is 'heavyer' than array but on the other hand it is easy to reconstruct a list so that you can throw away a part of it when it becomes obsolete. You could also use a wrapper which would only contain indexes saying which part of the list is up-to-date and which part may be reused.
This will help you improve performance, but will require a little bit more (reusable) memory.
Allocating by chunk and delete[]-ing and new[]-ing on the way seems like the good solution. It may be possible to do as little as memory management as possible. Do not reuse chunk yourself, simply deallocate old one and allocate new chunks when needed.
Related
I have a program, that uses dynamic programming to calculate some information. The problem is, that theoretically the used memory grows exponentially. Some filters that I use limit this space, but for a big input they also can't avoid that my program runs out of RAM - Memory.
The program is running on 4 threads. When I run it with a really big input I noticed, that at some point the program starts to use the swap memory, because my RAM is not big enough. The consequence of this is, that my CPU-usage decreases from about 380% to 15% or lower.
There is only one variable that uses the memory which is the following datastructure:
Edit (added type) with CLN library:
class My_Map {
typedef std::pair<double,short> key;
typedef cln::cl_I value;
public:
tbb::concurrent_hash_map<key,value>* map;
My_Map() { map = new tbb::concurrent_hash_map<myType>(); }
~My_Map() { delete map; }
//some functions for operations on the map
};
In my main program I am using this datastructure as globale variable:
My_Map* container = new My_Map();
Question:
Is there a way to avoid the shifting of memory between SWAP and RAM? I thought pushing all the memory to the Heap would help, but it seems not to. So I don't know if it is possible to maybe fully use the swap memory or something else. Just this shifting of memory cost much time. The CPU usage decreases dramatically.
If you have 1 Gig of RAM and you have a program that uses up 2 Gb RAM, then you're going to have to find somewhere else to store the excess data.. obviously. The default OS way is to swap but the alternative is to manage your own 'swapping' by using a memory-mapped file.
You open a file and allocate a virtual memory block in it, then you bring pages of the file into RAM to work on. The OS manages this for you for the most part, but you should think about your memory usage so not to try to keep access to the same blocks while they're in memory if you can.
On Windows you use CreateFileMapping(), on Linux you use mmap(), on Mac you use mmap().
The OS is working properly - it doesn't distinguish between stack and heap when swapping - it pages you whatever you seem not to be using and loads whatever you ask for.
There are a few things you could try:
consider whether myType can be made smaller - e.g. using int8_t or even width-appropriate bitfields instead of int, using pointers to pooled strings instead of worst-case-length character arrays, use offsets into arrays where they're smaller than pointers etc.. If you show us the type maybe we can suggest things.
think about your paging - if you have many objects on one memory page (likely 4k) they will need to stay in memory if any one of them is being used, so try to get objects that will be used around the same time onto the same memory page - this may involve hashing to small arrays of related myType objects, or even moving all your data into a packed array if possible (binary searching can be pretty quick anyway). Naively used hash tables tend to flay memory because similar objects are put in completely unrelated buckets.
serialisation/deserialisation with compression is a possibility: instead of letting the OS swap out full myType memory, you may be able to proactively serialise them into a more compact form then deserialise them only when needed
consider whether you need to process all the data simultaneously... if you can batch up the work in such a way that you get all "group A" out of the way using less memory then you can move on to "group B"
UPDATE now you've posted your actual data types...
Sadly, using short might not help much because sizeof key needs to be 16 anyway for alignment of the double; if you don't need the precision, you could consider float? Another option would be to create an array of separate maps...
tbb::concurrent_hash_map<double,value> map[65536];
You can then index to map[my_short][my_double]. It could be better or worse, but is easy to try so you might as well benchmark....
For cl_I a 2-minute dig suggests the data's stored in a union - presumably word is used for small values and one of the pointers when necessary... that looks like a pretty good design - hard to improve on.
If numbers tend to repeat a lot (a big if) you could experiment with e.g. keeping a registry of big cl_Is with a bi-directional mapping to packed integer ids which you'd store in My_Map::map - fussy though. To explain, say you get 987123498723489 - you push_back it on a vector<cl_I>, then in a hash_map<cl_I, int> set [987123498723489 to that index (i.e. vector.size() - 1). Keep going as new numbers are encountered. You can always map from an int id back to a cl_I using direct indexing in the vector, and the other way is an O(1) amortised hash table lookup.
I am a student of a system software faculty. Now I'm developing a memory manager for Windows. Here's my simple implementation of malloc() and free():
HANDLE heap = HeapCreate(0, 0, 0);
void* hmalloc(size_t size)
{
return HeapAlloc(heap, 0, size);
}
void hfree(void* memory)
{
HeapFree(heap, 0, memory);
}
int main()
{
int* ptr1 = (int*)hmalloc(100*sizeof(int));
int* ptr2 = (int*)hmalloc(100*sizeof(int));
int* ptr3 = (int*)hmalloc(100*sizeof(int));
hfree(ptr2);
hfree(ptr3);
hfree(ptr1);
return 0;
}
It works fine. But I can't understand is there a reason to use multiple heaps? Well, I can allocate memory in the heap and get the address to an allocated memory chunk. But here I use ONE heap. Is there a reason to use multiple heaps? Maybe for multi-threaded/multi-process applications? Please explain.
The main reason for using multiple heaps/custom allocators are for better memory control. Usually after lots of new/delete's the memory can get fragmented and loose performance for the application (also the app will consume more memory). Using the memory in a more controlled environment can reduce heap fragmentation.
Also another usage is for preventing memory leaks in the application, you could just free the entire heap you allocated and you don't need to bother with freeing all the object allocated there.
Another usage is for tightly allocated objects, if you have for example a list then you could allocate all the nodes in a smaller dedicated heap and the app will gain performance because there will be less cache misses when iterating the nodes.
Edit: memory management is however a hard topic and in some cases it is not done right. Andrei Alexandrescu had a talk at one point and he said that for some application replacing the custom allocator with the default one increased the performance of the application.
This is a good link that elaborates on why you may need multiple heap:
https://caligari.dartmouth.edu/doc/ibmcxx/en_US/doc/libref/concepts/cumemmng.htm
"Why Use Multiple Heaps?
Using a single runtime heap is fine for most programs. However, using multiple
heaps can be more efficient and can help you improve your program's performance
and reduce wasted memory for a number of reasons:
1- When you allocate from a single heap, you may end up with memory blocks on
different pages of memory. For example, you might have a linked list that
allocates memory each time you add a node to the list. If you allocate memory for
other data in between adding nodes, the memory blocks for the nodes could end up
on many different pages. To access the data in the list, the system may have to
swap many pages, which can significantly slow your program.
With multiple heaps, you can specify which heap you allocate from. For example,
you might create a heap specifically for the linked list. The list's memory blocks
and the data they contain would remain close together on fewer pages, reducing the
amount of swapping required.
2- In multithread applications, only one thread can access the heap at a time to
ensure memory is safely allocated and freed. For example, say thread 1 is
allocating memory, and thread 2 has a call to free. Thread 2 must wait until
thread 1 has finished its allocation before it can access the heap. Again, this
can slow down performance, especially if your program does a lot of memory
operations.
If you create a separate heap for each thread, you can allocate from them
concurrently, eliminating both the waiting period and the overhead required to
serialize access to the heap.
3- With a single heap, you must explicitly free each block that you allocate. If you
have a linked list that allocates memory for each node, you have to traverse the
entire list and free each block individually, which can take some time.
If you create a separate heap for that linked list, you can destroy it with a
single call and free all the memory at once.
4- When you have only one heap, all components share it (including the IBM C and
C++ Compilers runtime library, vendor libraries, and your own code). If one
component corrupts the heap, another component might fail. You may have trouble
discovering the cause of the problem and where the heap was damaged.
With multiple heaps, you can create a separate heap for each component, so if
one damages the heap (for example, by using a freed pointer), the others can
continue unaffected. You also know where to look to correct the problem."
A reason would be the scenario that you need to execute a program internally e.g. running simulation code. By creating your own heap you could allow that heap to have execution rights which by default for security reasons is turned off. (Windows)
You have some good thoughts and this'd work for C but in C++ you have destructors, it is VERY important they run.
You can think of all types as having constructors/destructors, just that logically "do nothing".
This is about allocators. See "The buddy algorithm" which uses powers of two to align and re-use stuff.
If I allocate 4 bytes somewhere, my allocator might allocate a 4kb section just for 4 byte allocations. That way I can fit 1024 4 byte things in the block, if I need more add another block and so forth.
Ask it for 4kb and it wont allocate that in the 4byte block, it might have a separate one for larger requests.
This means you can keep big things together. If I go 17 bytes then 13 bytes the 1 byte and the 13byte gets freed, I can only stick something in there of <=13 bytes.
Hence the buddy system and powers of 2, easy to do using lshifts, if I want a 2.5kb block, I allocate it as the smallest power of 2 that'll fit (4kb in this case) that way I can use the slot afterwards for <=4kb items.
This is not for garbage collection, this is just keeping things more compact and neat, using your own allocator can stop calls to the OS (depending on the default implementation of new and delete they might already do this for your compiler) and make new/delete very quick.
Heap-compacting is very different, you need a list of every pointer that points to your heap, or some way to traverse the entire memory graph (like spits Java) so when you move stuff round and "compact" it you can update everything that pointed to that thing to where it currently is.
The only time I ever used more than one heap was when I wrote a program that would build a complicated data structure. It would have been non-trivial to free the data structure by walking through it and freeing the individual nodes, but luckily for me the program only needed the data structure temporarily (while it performed a particular operation), so I used a separate heap for the data structure so that when I no longer needed it, I could free it with one call to HeapDestroy.
The problem is following: I have certain amount of words (let's say 20M), each containing some bits used as flags; all stored in single continuous binary file.
What I would like to do is to get access to those words in container like style, so container_instance[i] allows me to access i-th word. To get things more complicated, I cannot store all words in memory at one time, they have to be stored back to file and memory freed for those not used for long period. To simplify things the whole sequence is partitioned to 1K fragments, so we need to free and allocate such 1K blocks. Memory should be freed after some time or after certain number of times container have been accessed.
Thread safety in nice to have. But I can protect externally.
The implementation I have currently only allocates blocks on demand (empty or read from file if they are available; file is not sparse, so everything after the last byte in file is allocated empty) and it is not nicely done. Not frees at all, so unused blocks remain in memory forever.
I started to think about nice looking solution and I would like to know whether any elements from STL or Boosts can help me build such container not by engraving it step by step from scratch?
I am not expecting full solutions, rather pointing "you can use that for that".
You can use mmap system call to map your file into memory. You can use pointer arithmetic with that buffer, so access by index is not a trouble.
Mapped pages are virutual and managed by the kernel, allowing to save unused memory blocks and load/flush them at transparently to you. Also, using madvise probably can enable some optimisations.
memcpy/memmove duplicate (copy the data) from source to destination. Does anything exist to move pages from one virtual address to another without doing an actual byte by byte copy of the source data? It seems to be perfectly possible to me, but does any operating system actually allow this? It seems odd to me that dynamic arrays are such a widespread and popular concept but that growing them by physically copying is such a wasteful operation. It just doesn't scale when you start talking about array sizes in the gigabytes (e.g. imagine growing a 100GB array into a 200GB array. That's a problem that's entirely possible on servers in the < $10K range now.
void* very_large_buffer = VirtualAlloc(NULL, 2GB, MEM_COMMIT);
// Populate very_large_buffer, run out of space.
// Allocate buffer twice as large, but don't actually allocate
// physical memory, just reserve the address space.
void* even_bigger_buffer = VirtualAlloc(NULL, 4GB, MEM_RESERVE);
// Remap the physical memory from very_large_buffer to even_bigger_buffer without copying
// (i.e. don't copy 2GB of data, just copy the mapping of virtual pages to physical pages)
// Does any OS provide support for an operation like this?
MoveMemory(very_large_buffer, even_bigger_buffer, 2GB)
// Now very_large_buffer no longer has any physical memory pages associated with it
VirtualFree(very_large_buffer)
To some extent, you can do that with mremap on Linux.
That call plays with the process's page table to do a zero-copy reallocation if it can. It is not possible in all cases (address space fragmentation, and simply the presence of other existing mappings are an issue).
The man page actually says this:
mremap() changes the mapping between virtual addresses and memory pages. This can be used to implement a very efficient realloc(3).
Yes it's a common use of memory mapped files to 'move' or copy memory between process by mapping different views of the file
Every POSIX system is able to do this. If you use mmap with a file descriptor (obtained by open or shm_open) and not anonymously you can unmap it, then truncate (shrink or grow) and then map it again. You may and often will get a different virtual address for the same pages.
I mean, you'd never be able to absolutely guarantee that there would be no active memory in that next 100GB, so you might not be able to make it contiguous.
On the other hand, you could use a ragged array (an array of arrays) where the arrays do not have to be next to each other (or even the same size). Many of the advantages of dynamic arrays may not scale to the 100GB realm.
I need to set some default value for all entires in a very large array.
It takes me quite long time (110-120 ms) and i suspect it happens because of misses in memory.
I use memset/std:fill to set the default value. Is there a way to make sure that the array will reside in memory before the memset/fill?
Assuming this is a large memory-mapped file, you can use the madvise() libc call with the MADV_WILLNEED argument to hint to the OS that you'll be wanting to access the region mentioned soon.
However YMMV, as the array needs to be large enough that the benefit of the resulting syscall isn't outweighed by the cost of making the call.
You can lock memory at per-page granuality using mlock, though only up to a fixed amount (I'm not sure what the limit is on OS X, but you can check it using getrlimit with RLIMIT_MEMLOCK).
Most likely you have a multiple core processor and functions like memset actually degrade in performance when not used on single core CPUs. It's possible that mutex locking are causing the slowdown. Try allocating memory on the stack instead of dynamic memory. Since it's a very large array then I would experiment making my own memory manager and store segments of it in multiple threads (but that's just an idea I had after reading an article fast). A standard way of doing it would be to use one memory allocator per thread. In any case I would look into something else than memset.
Maybe the following aticle would help