I'm in the process of switching over to Julia from other programming languages and one of the things that Julia will let you hang yourself on is memory. I think this is likely a good thing, a programming language where you actually have to think about some amount of memory management forces the coder to write more efficient code. This would be in contrast to something like R where you can seemingly load datasets that are larger than the allocated memory. Of course, you can't actually do that, so I wonder how does R get around that problem?
Part of what I've done in other programming languages is work on large tabular datasets, often converted over to a R dataframe or a matrix. I think the way this is handled in Julia is to stream data in wherever possible, so my main question is this:
Is it better to use readline("my_file.txt") to access data or is it better to use open("my_file.txt", "w")? If possible, wouldn't it be better to access a large dataset all at once for speed? Or would it be better to always stream data?
I hope this makes sense. Any further resources would be greatly appreciated.
I'm not an extensive user of Julia's data-ecosystem packages, but CSV.jl offers the Chunks and Rows alternatives to File, and these might let you process the files incrementally.
While it may not be relevant to your use case, the mechanisms mentioned in #Przemyslaw Szufel's answer are used other places as well. Two I'm familiar with are the TiffImages.jl and NRRD.jl packages, both I/O packages mostly for loading image data into Julia. With these, you can load terabyte-sized datasets on a laptop. There may be more packages that use the same mechanism, and many package maintainers would probably be grateful to receive a pull request that supports optional memory-mapping when applicable.
In R you cannot have a data frame larger than memory. There is no magical buffering mechanism. However, when running R-based analytics you could use a disk.frame package for that.
Similarly, in Julia if you want to process data frames larger than memory you need to use am appropriate package. The most reasonable and natural option in Julia ecosystem is JuliaDB.
If you want to do something more low-level solution have a look at:
Mmap that provides Memory-mapped I/O that exactly solves the issue of conveniently handling data too large to fit into memory
SharedArrays that offers a disk mapped array with implementation based on Mmap.
In conclusion, if your data is data frame based - try JuliaDB, otherwise have a look at Mmap and SharedArrays (look at the filename parameter)
Related
I normally resolve this by allocating more memory, however sometimes I find mys elf allocating quite a lot and I wonder if there are other alternatives I can use to solve this, is there a way to make the map jobs smaller, perhaps in aggregate they use same memory, but not all the time
I am mostly trying to learn. I can provide an example but I am not asking about a query or a specific instance, more in general
It isn't clear to me when it's a good idea to use VK_IMAGE_LAYOUT_GENERAL as opposed to transitioning to the optimal layout for whatever action I'm about to perform. Currently, my policy is to always transition to the optimal layout.
But VK_IMAGE_LAYOUT_GENERAL exists. Maybe I should be using it when I'm only going to use a given layout for a short period of time.
For example, right now, I'm writing code to generate mipmaps using vkCmdBlitImage. As I loop through the sub-resources performing the vkCmdBlitImage commands, should I transition to VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL as I scale down into a mip, then transition to VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL when I'll be the source for the next mip before finally transitioning to VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL when I'm all done? It seems like a lot of transitioning, and maybe generating the mips in VK_IMAGE_LAYOUT_GENERAL is better.
I appreciate the answer might be to measure, but it's hard to measure on all my target GPUs (especially because I haven't got anything running on Android yet) so if anyone has any decent rule of thumb to apply it would be much appreciated.
FWIW, I'm writing Vulkan code that will run on desktop GPUs and Android, but I'm mainly concerned about performance on the latter.
You would use it when:
You are lazy
You need to map the memory to host (unless you can use PREINITIALIZED)
When you use the image as multiple incompatible attachments and you have no choice
For Store Images
( 5. Other cases when you would switch layouts too much (and you don't even need barriers) relatively to the work done on the images. Measurement needed to confirm GENERAL is better in that case. Most likely a premature optimalization even then.
)
PS: You could transition all the mip-maps together to TRANSFER_DST by a single command beforehand and then only the one you need to SRC. With a decent HDD, it should be even best to already have them stored with mip-maps, if that's a option (and perhaps even have a better quality using some sophisticated algorithm).
PS2: Too bad, there's not a mip-map creation command. The cmdBlit most likely does it anyway under the hood for Images smaller than half resolution....
If you read from mipmap[n] image for creating the mipmap[n+1] image then you should use the transfer image flags if you want your code to run on all Vulkan implementations and get the most performance across all implementations as the flags may be used by the GPU to optimize the image for reads or writes.
So if you want to go cross-vendor only use VK_IMAGE_LAYOUT_GENERAL for setting up the descriptor that uses the final image and not image reads or writes.
If you don't want to use that many transitions you may copy from a buffer instead of an image, though you obviously wouldn't get the format conversion, scaling and filtering that vkCmdBlitImage does for you for free.
Also don't forget to check if the target format actually supports the BLIT_SRC or BLIT_DST bits. This is independent of whether you use the transfer or general layout for copies.
I have read several times that turning on compression in HDF5 can lead to better read/write performance.
I wonder what ideal settings can be to achieve good read/write performance at:
data_df.to_hdf(..., format='fixed', complib=..., complevel=..., chunksize=...)
I'm already using fixed format (i.e. h5py) as it's faster than table. I have strong processors and do not care much about disk space.
I often store DataFrames of float64 and str types in files of approx. 2500 rows x 9000 columns.
There are a couple of possible compression filters that you could use.
Since HDF5 version 1.8.11 you can easily register a 3rd party compression filters.
Regarding performance:
It probably depends on your access pattern because you probably want to define proper dimensions for your chunks so that it aligns well with your access pattern otherwise your performance will suffer a lot. For example if you know that you usually access one column and all rows you should define your chunk shape accordingly (1,9000). See here, here and here for some infos.
However AFAIK pandas usually will end up loading the entire HDF5 file into memory unless you use read_table and an iterator (see here) or do the partial IO yourself (see here) and thus doesn't really benefit that much of defining a good chunk size.
Nevertheless you might still benefit from compression because loading the compressed data to memory and decompressing it using the CPU is probably faster than loading the uncompressed data.
Regarding your original question:
I would recommend to take a look at Blosc. It is a multi-threaded meta-compressor library that supports various different compression filters:
BloscLZ: internal default compressor, heavily based on FastLZ.
LZ4: a compact, very popular and fast compressor.
LZ4HC: a tweaked version of LZ4, produces better compression ratios at the expense of speed.
Snappy: a popular compressor used in many places.
Zlib: a classic; somewhat slower than the previous ones, but achieving better compression ratios.
These have different strengths and the best thing is to try and benchmark them with your data and see which works best.
I have this molecular dynamics program that writes atom position and velocities to a file at every n steps of simulation. The actual writing is taking like 90% of the running time! (checked by eiminating the writes) So I desperately need to optimize that.
I see that some fortrans have an extension to change the write buffer size (called i/o block size) and the "number of blocks" at the OPEN statement, but it appears that gfortran doesn't. Also I read somewhere that gfortran uses 8192 bytes write buffer.
I even tried to do an FSTAT (right after opening, is that right?) to see what is the block size and number of blocks it is using but it returns -1 on both. (compiling for windows 64 bit)
Isn't there a way to enlarge the write buffer for a file in gfortran? Will it be diferent compiling for linux than for windows?
I'd really really rather stay in fortran but as a desperate measure isn't there a way to do so by adding some c routine?
thanks!
IanH question is key. Unformatted IO is MUCH faster than formatted. The conversion from base 2 to base 10 is very CPU intensive. If you don't need the values to be human readable, then use unformatted IO. If you want to be able to read the values in another language, then use access='stream'.
Another approach would be to add your own buffering. Replace the write statement with a call to a subroutine. Have that subroutine store values and write only when it has received M values. You'll also have to have a "flush" call to the subroutine to cause it to write the last values, if they are fewer them M.
If gcc C is faster at IO, you could mix Fortran and C with Fortran's ISO_C_Binding: https://stackoverflow.com/questions/tagged/fortran-iso-c-binding. There are examples of the use of the ISO C Binding in the gfortran manual under "Mixed Language Programming".
If you spend 90% of your runtime writing coords/vels every n timesteps, the obvious quick fix would be to instead write data every, say, n/100 timestep. But I'm sure you already thought of that yourself.
But yes, gfortran has a fixed 8k buffer, whose size cannot be changed except by modifying the libgfortran source and rebuilding it. The reason for the buffering is to amortize the syscall overhead; (simplistic) tests on Linux showed that 8k is sufficient and more than that goes far into diminishing returns territory. That being said, if you have some substantiated claims that bigger buffers are useful on some I/O patterns and/or OS, there's no reason why the buffer can't be made larger in a future release.
As for you performance issues, as already mentioned, unformatted is a lot faster than formatted I/O. Additionally, gfortran has rather high per-IO-statement overhead. You can amortize that by writing arrays (or, array sections) rather than individual elements (this matters mostly for unformatted, for formatted IO there is so much to do that this doesn't help that much).
I am thinking that if cost of IO is comparable or even larger than the effort of simulation, then it probably isn't such a good idea to store all these data to disk the first place. It is better to do whatever processing you intend to do directly during the simulation, instead of saving lots of intermediate data them later read them in again to do the processing.
Moreover, MD is an inherently highly parallelizable problem, and with IO you will severely cripple the efficiency of parallelization! I would avoid IO whenever possible.
For individual trajectories, normally you just need to store the initial condition of each trajectory, along with its key statistics, or important snapshots at a small number of time values. When you need one specific trajectory plotted you can regenerate the exact same trajectory or section of trajectory from the initial condition or the closest snapshot, and with similar cost as reading it from the disk.
Suppose I want to do following opeartions on my 2-core machine:
Read a very large file
Compute
Does the file reading operation need to consume 1 core? Previously I just create 2 threads, one to read file and one to compute? Should I create an additional thread to do compute?
Thanks.
Edit
Thanks guys, yea, we should always consider if the file I/O blocks the computing. Now let's just consider that the file I/O will never block computing, you can think the computing doesn't depends on the file's data, we just read the file in for future processing. Now we have 2 core, we need to read in a file, and we need to do computing, is it the best solution to create 3 threads, 1 for file reading and 2 for computing, as most of you has already pointed out: file reading consumes very little CPU?
It depends on how your hardware is configured. Normally, reading is not CPU-intensive, thanks to DMA. It may be very expensive though, if it initiates swap-out of other applications. But there is more to it.
Don't read a huge file at once if you can
If your file is really big, you should use mmap or sequential processing, when you don't need to read a whole file at once. Try to consume it by chunks is possible.
For example, to sum all values in a huge file, you don't need to load this file into the memory. You can process it by small chunks, accumulating the sum. Memory is an expensive resource in most situations.
Reading is sequential
Does the file reading operation need to consume 1 core?
Yes, I think most low-level read operations are implemented sequentially (consume 1 core).
You can avoid blocking on read operation if you use asynchronous I/O, but it is just a variation of the same "read by small chunks" technique. You can launch several small asynchronous read operations at once, but you have always to check if an operation has finished before you use the result.
See also this Stack Overflow answer to a related question).
Reading and computing in parallel
Previously I just create 2 threads, one to read file and one to compute? Should I create an additional thread to do compute?
It depends, if you need all data to start computations, than there is no reason to start computation in parallel. It will have to wait effectively until reading is done.
If you can start computing even with partial data, likely you don't need to read the whole file at once. And it is usually much better not to do so with huge files.
What is your bottleneck — computation or IO?
Finally, you should know if your task is computation-bound or input-output bound. If it is limited by the performance of input-output subsystem, there is little benefit in parallelizing computation. If computation is very CPU-intensive, and reading time is negligible, you can benefit from parallelizing computation. Input-output is usually a bottleneck unless you are doing some number-crunching.
This is a good candidate for parallelization, because you have two types of operations here - disk I/O (for reading the file), and CPU load (for your computations). So the first step would be to write your application such that the file I/O wasn't blocking the computation. You could do this by reading a little bit at a time from the file and handing it off to the compute thread.
But now you're saying you have two cores that you want to utilize. Your second thought about parallelizing the CPU-intensive part is correct, because we can only parallelize compute tasks if we have more than one processor to use. But, it might be the case that the blocking part of your application is still the file I/O - that depends on a lot of factors, and the only way to tell what level of parallelization is appropriate is to benchmark.
SO required caveat: multithreading is hard and error-prone, and it's better to have correct code than fast code, if you can pick only one. But I don't advocate against threads, as you may find from others on the site.
I would think this depends on the computation you are performing. If you are doing very heavy computations then I would suggest threading the application. Reading a file demands very little from your CPU and because of this, the overhead created by threading the application might slow it down.
Another thing to consider is if you need to load the entire file before you can compute, if so, there is no point in threading it at all as you will have to complete one action before you can perform the other.