Slow write compared to raw writes (>= 10GB)

Hi All

I am writing a streaming storage application and would really like to utilize the HDF5 capabilities in storing meta-data. I’m also receiving a significant amount of data, and would like to stream 5+ GB/s continuously. I only write, never read until the file is closed for writing first.

When I write a simple file storage function with 400MB chucks, I get ~600 MB/s write speed. If I compare this to a normal WriteFile call in Windows, I get ~2’000 MB/s. I only need the ‘FILE_FLAG_WRITE_THROUGH’ file creation.

How can I get the same write-performance for the HDF-5 format, as I can for this? I can easily store 1 second of data before writing, if this would help, but I don’t see any performance improvement in increasing the chunk-size.

Linux/Windows, I see the same performance numbers on both OS’s.

I will try to attach a minimum working example in a bit.

See my comment on Write perfomance of raw data

Use H5Pset_fill_time(..., H5D_FILL_TIME_NEVER) to avoid writing data twice.

If you want to be as close as possible to pwrite, use H5Dwrite_chunk instead of H5Dwrite.


1 Like

Benchmark has an example for cross product test, see below, please post the out put on a recent linux/posix system if your output is significantly different. For larger blocks you should expect IO throughput in the 70-95% of the underlying IO bandwidth, For small chunks you have to the a lot more, as Gerd mentioned direct chunk IO is the way to go. Then you roll your packing algorithm for various sizes, and build your own filter chain. Or you could just use h5::append

(The test was performed on my Lenovo X1 couple of years ago. )

steven@io:~/projects/h5bench/examples/capi$ make
g++ -I/usr/local/include -I/usr/include -I../../include -o capi-test.o   -std=c++17 -Wno-attributes -c capi-test.cpp
g++ capi-test.o -lhdf5  -lz -ldl -lm -o capi-test
taskset 0x1 ./capi-test
[name                                              ][total events][Mi events/s] [ms runtime / stddev] [    MiB/s / stddev ]
fixed length string CAPI                                    10000     625.0000         0.02     0.000   24461.70     256.9
fixed length string CAPI                                   100000     122.7898         0.81     0.038    4917.70     213.3
fixed length string CAPI                                  1000000      80.4531        12.43     0.217    3218.60      56.6
fixed length string CAPI                                 10000000      79.7568       125.38     0.140    3189.80       3.6
rm capi-test.o
int main(int argc, const char **argv){
  size_t max_size = *std::max_element(record_size.begin(), record_size.end());

  h5::fd_t fd = h5::create("h5cpp.h5", H5F_ACC_TRUNC);
  auto strings = h5::utils::get_test_data<std::string>(max_size, 10, sizeof(fl_string_t));
	std::vector<char[sizeof(fl_string_t)]> data(strings.size());
		for (size_t i = 0; i < data.size(); i++)
			strncpy(data[i], strings[i].data(), sizeof(fl_string_t));
  // set the transfer size for each batch
  std::vector<size_t> transfer_size;
  for (auto i : record_size)
      transfer_size.push_back(i * sizeof(fl_string_t));
  //use H5CPP  modify VL type to fixed length
  h5::dt_t<fl_string_t> dt{H5Tcreate(H5T_STRING, sizeof(fl_string_t))};
  H5Tset_cset(dt, H5T_CSET_UTF8);

  std::vector<h5::ds_t> ds;
  // create separate dataset for each batch
  for(auto size: record_size) ds.push_back(
    h5::create<fl_string_t>(fd, fmt::format("fixe length string CAPI-{:010d}", size), 
    chunk_size, h5::current_dims{size}, dt));

  // EXPERIMENT: arguments, including lambda function may be passed in arbitrary order
    bh::name{"fixed length string CAPI"}, record_size, warmup, sample,
    [&](size_t idx, size_t size_) -> double {
        hsize_t size = size_;
        // memory space
        h5::sp_t mem_space{H5Screate_simple(1, &size, nullptr )};
        // file space
        h5::sp_t file_space{H5Dget_space(ds[idx])};
        // IO call
        H5Dwrite( ds[idx], dt, mem_space, file_space, H5P_DEFAULT,;
        return transfer_size[idx];

Thank you both for your reply. I’ll look in the referenced post as it is the same issue we are facing.