This tool and API implements parallel block gzip.
For many formats, in particular Genomics Data Formats, data are compressed in fixed-length blocks such that they can be easily indexed based on a (genomic) coordinate order, since typically each block is sorted according to this order.
This allows for each block to be individually compressed (deflated), or more importantly, decompressed (inflated), with the latter enabling random retrieval of data in large files (gigabytes to terabytes).
pbgzip is not limited to any particular format, but certain features are tailored to Genomics Data Formats when enabled (see below).
Parallel decompression is somewhat faster, but truly the speedup comes during compression.
Author: Nils Homer
- Compile PBGZIP:
sh autogen.sh && ./configure && make- Install
make installigzip is a high-performance compression library for gzip or deflate compression for Genome data.
For BAM (or SAM) formats (see the SAM/BAM Specification), it can significantly speed up the compression, with minimal loss in compression ratio when compared to the fastest level of zlib.
In pbgzip, we can utilize the BAM-specific speedup, so using igzip in this context may not yield good results for non-BAM data.
The BAM-specific igzip libraries must installed prior to compiling pbgzip.
We have included the igzip source ready for installing the bam-specific igzip libraries.
Please compile and install igzip:
cd igzip
make
make install
cd ..
To override the install path, set the INSTALL_PATH environment variable before compiling and installing igzip.
You may also need to updated your LD_LIBRARY_PATH variable to include the INSTALL_PATH:
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib
If the compilation fails, you may be on a system that is not supported by igzip; please contact the original authors for igzip support.
Make sure to enable igzip when running configure:
./configure --enable-igzip
Please thank the original authors of igzip: Jim Guilford, Vinodh Gopal, Sean Gulley and Wajdi Feghali.
Also thank Paolo Narvaez, Mishali Naik and Gil Wolrich for their contributions.
Finally, thank Intel for sponsoring the igzip work.
Intel's Integrated Performance Primitives (Intel IPP) can boost performance of compression (deflation) by 20-30%. Currently, Intel IPP 9.0 is supported.
Make sure to enable ipp when running configure:
./configure --enable-ipp
-
Install Intel's IPP (see Intel IPP]).
-
Make sure that the path to the installed Intel IPP libraries is either (a)added to your relevant environment variables, or (b) or specified during the configure command.
(a) Add the following to your .bashrc. Please modify the path for your installation.
export LIBRARY_PATH=${LIBRARY_PATH}:/opt/intel/compilers_and_libraries_2016.0.083/mac/ipp/lib
export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/opt/intel/compilers_and_libraries_2016.0.083/mac/ipp/lib
(b) Append the following to the arguments to configure. Please modify the path for your installation.
LDFLAGS=-L/opt/intel/compilers_and_libraries_2016.0.083/mac/ipp/lib CFLAGS=-I/opt/intel/compilers_and_libraries_2016.0.083/mac/ipp/include
For developer issues, see: https://github.com/nh13/pbgzip/issues
Due to the requirement that we must fit a chunk of compressed data into a maximum size block, we may fail at compression when the data has sufficiently high entropy (is too random). For example, try compressing a very random set of data:
head -c 10M </dev/urandom | pbgzip -c > myfile.gz
The compression should fail.
Fortunately, bgzip should succeed on this data, but pbgzip currently does not support this case.
Developer Note: we would need to keep a buffer of data that was fed to consumers, such that we rewind back to try to compress fewer bytes in the given block that failed deflation.
In same cases, you may get: undefined reference to `rpl_malloc'. This is due to not being able to find the malloc function.
Try the following:
ac_cv_func_malloc_0_nonnull=yes ./configure <your configure options>