Intel® QuickAssist Technology (QAT) QATzip Library

Table of Contents

• Introduction
• Licensing
• Features
• Hardware Requirements
• Software Requirements
• Additional Information
• Limitations
• Installation Instructions
  • Install with the in-tree QAT package
  • Install with the out-of-tree QAT package
  • Configuration
  • Enable qzstd
• Test QATzip
• Performance Test With QATzip
• QATzip API manual
• Intended Audience
• Open Issues
• Legal

Introduction

QATzip is a user space library which builds on top of the Intel® QuickAssist Technology user space library, to provide extended accelerated compression and decompression services by offloading the actual compression and decompression request(s) to the Intel® Chipset Series. QATzip produces data using the standard gzip* format (RFC1952) with extended headers or lz4* blocks with lz4* frame format. The data can be decompressed with a compliant gzip* or lz4* implementation. QATzip is designed to take full advantage of the performance provided by Intel® QuickAssist Technology.

The currently supported formats include:

Data Format	Algorithm	QAT device	Description
QZ_DEFLATE_4B	deflate*	QAT 1.x and QAT 2.0	Data is in DEFLATE* with a 4 byte header
QZ_DEFLATE_GZIP	deflate*	QAT 1.x and QAT 2.0	Data is in DEFLATE* wrapped by Gzip* header and footer
QZ_DEFLATE_GZIP_EXT	deflate*	QAT 1.x and QAT 2.0	Data is in DEFLATE* wrapped by Intel® QAT Gzip* extension header and footer
QZ_DEFLATE_RAW	deflate*	QAT 1.x and QAT 2.0	Data is in raw DEFLATE* without any additional header. (Only support compression, decompression will fallback to software)
QZ_LZ4	lz4*	QAT 2.0	Data is in LZ4* wrapped by lz4* frame
QZ_LZ4S	lz4s*	QAT 2.0	Data is in LZ4S* blocks

Licensing

The Licensing of the files within this project is split as follows:

Intel® Quickassist Technology (QAT) QATzip - BSD License. Please see the LICENSE file contained in the top level folder. Further details can be found in the file headers of the relevant files.

Example Intel® Quickassist Technology Driver Configuration Files contained within the folder hierarchy config_file - Dual BSD/GPLv2 License. Please see the file headers of the configuration files, and the full GPLv2 license contained in the file LICENSE.GPL within the config_file folder.

Features

• Acceleration of compression and decompression utilizing Intel® QuickAssist Technology, including a utility to compress and decompress files.

• Dynamic memory allocation for zero copy, by exposing qzMalloc() and qzFree() allowing working buffers to be pinned, contiguous buffers that can be used for DMA operations to and from the hardware.

• Instance over-subscription, allowing a number of threads in the same process to seamlessly share a smaller number of hardware instances.

• Memory allocation backed by huge page and kernel memory to provide access to pinned, contiguous memory. Allocating from huge-page when kernel memory contention.

• Configurable accelerator device sharing among processes.

• Optional software failover for both compression and decompression services. QATzip may switch to software if there is insufficient system resources including acceleration instances or memory. This feature allows for a common software stack between server platforms that have acceleration devices and non-accelerated platforms.

• Provide streaming interface of compression and decompression to achieve better compression ratio and throughput for data sets that are submitted piecemeal.

• 'qzip' utility supports compression from regular file, pipeline and block device.

• For QATzip GZIP* format, try hardware decompression first before switching to software decompression.

• Enable adaptive polling mechanism to save CPU usage in stress mode.

• 'qzip' utility supports compression files and directories into 7z format.

• Support QATzip Gzip* format, it includes 10 bytes header and 8 bytes footer:

  | ID1 (1B) | ID2(0x8B) (1B) | Compression Method (8 = DEFLATE*) (1B) | Flags (1B) | Modification Time (4B) | Extra Flags (1B) | OS (1B) | Deflate Block| CRC32(4B)| ISIZE(4B)|

• Support QATzip Gzip* extended format. This consists of the standard 10 byte Gzip* header and follows RFC 1952 to extend the header by an additional 14 bytes. The extended headers structure is below:

  | Length of ext. header (2B) | SI1('Q') (1B) | SI2('Z') (1B) | Length of subheader (2B) | Intel(R) defined field 'Chunksize' (4B) | Intel(R) defined field 'Blocksize' (4B) |

• Support Intel® QATzip 4 byte header, the header indicates the length of the compressed block followed by the header.

  | Intel(R) defined Header (4B)|deflate\* block|

• Support QATzip lz4* format. This format is structured as follows:

  | MagicNb(4B) |FLG(1B)|BD(1B)| CS(8B)|HC(1B)| |lz4\* Block | EndMark(4B)|

Hardware Requirements

This QATzip library supports compression and decompression offload to the following acceleration devices:

• Intel® C62X Series Chipset
• Intel® Communications Chipset 8925 to 8955 Series
• Intel® Communications Chipset 8960 to 8970 Series
• Intel® C3XXX Series Chipset
• Intel® 4XXX Series

Software Requirements

This release was validated on the following:

• QATzip has been tested with the latest Intel® QuickAssist Acceleration Driver. Please download the QAT driver from the link Intel® QuickAssist Technology
• QATzip has been tested by Intel® on CentOS* 7.8.2003 with kernel 3.10.0-1127.19.1.el7.x86_64
• Zlib* library of version 1.2.7 or higher
• Suggest GCC* of version 4.8.5 or higher
• lz4* library of version 1.8.3 or higher
• zstd* library of version 1.5.0 or higher

Additional Information

• For QAT 1.x, the compression level in QATzip could be mapped to standard zlib* as below:

  • QATzip level 1 - 4, similar to zlib* level 1 - 4.
  • QATzip level 5 - 8, we map them to QATzip level 4.
  • QATzip level 9, we will use software zlib* to compress as level 9.

• For QAT 2.0, the compression level in QATzip could be mapped to standard zlib* or lz4* as below:

  • Will be updated in future releases.

• QATzip Compression Level Mapping:

  QATzip Level	QAT Level	QAT 2.0(deflate*, LZ4*, LZ4s*)	QAT1.7/1.8(Deflate*)
  1	CPA_DC_L1	2(HW_L1)	DEPTH_1
  2	CPA_DC_L2	2(HW_L1)	DEPTH_4
  3	CPA_DC_L3	2(HW_L1)	DEPTH_8
  4	CPA_DC_L4	2(HW_L1)	DEPTH_16
  5	CPA_DC_L5	2(HW_L1)	DEPTH_16
  6	CPA_DC_L6	8(HW_L6)	DEPTH_16
  7	CPA_DC_L7	8(HW_L6)	DEPTH_16
  8	CPA_DC_L8	8(HW_L6)	DEPTH_16
  9	CPA_DC_L9	16(HW_L9)	DEPTH_16
  10	CPA_DC_L10	16(HW_L9)	Unsupported
  11	CPA_DC_L11	16(HW_L9)	Unsupported
  12	CPA_DC_L12	16(HW_L9)	Unsupported

Limitations

• The partitioned internal chunk size of 16 KB is disabled, this chunk is used for QAT hardware DMA.
• For stream object, user should reset the stream object by calling qzEndStream() before reuse it in the other session.
• For stream object, user should clear stream object by calling qzEndStream() before clear session object with qzTeardownSession(). Otherwise, memory leak happens.
• For stream object, stream length must be smaller than strm_buff_sz, or QATzip would generate multiple deflate block in order and has the last block with BFIN set.
• For stream object, we will optimize the performance of the pre-allocation process using a thread-local stream buffer list in a future release.
• For 7z format, decompression only supports *.7z archives compressed by qzip.
• For 7z format, decompression only supports software.
• For 7z format, the header compression is not supported.
• For lz4* (de)compression, QATzip only supports 32KB history buffer.
• For zstd format compression, qzstd only supports hw_buffer_sz which is less than 128KB.
• Stream APIs only support "DEFLATE_GZIP", "DEFLATE_GZIP_EXT", "DEFLATE_RAW" for compression and "DEFLATE_GZIP", "DEFLATE_GZIP_EXT" for decompression now.

Installation Instructions

Install with the in-tree QAT package

• Please refer to link.

Install with the out-of-tree QAT package

1. The Installation of the out-of-tree QAT package refer to link.

Note

• If you run QAT as non-root user, more steps need to be manually applied, please refer to link.
  
• If SVM is not enabled, memory passed to QAT hardware must be DMA’able, Intel provides a USDM component which allocates/frees DMA-able memory. Please refer to link for USDM setting.

2. Install the package dependencies by running the below command:

sudo dnf install -y autoconf automake libtool zlib-devel lz4-devel
For Debian-based distros like Ubuntu, use these names for the latter two packages:
sudo apt -y install zlib1g-dev liblz4-dev

3. Configure the QATzip library by running the following commands:

cd QATzip/
export QZ_ROOT=`pwd`
export ICP_ROOT=/QAT/PACKAGE/PATH
./autogen.sh
./configure

Note
For more configure options, please run "./configure -h" for help.

4. Build and install the QATzip library by running the below commands:

make clean
make
sudo make install

Configuration

Note
This section is only required when you are using out-of-tree QAT package. if you are using qatlib with in-tree QAT package, please refer to link for details on configuring qatlib.

QAT programmer’s guide which provides information on the architecture of the software and usage guidelines, allows customization of runtime operation.

• Intel® QAT 1.7 linux Programmer's Guide
• Intel® QAT 2.0 linux Programmer's Guide

The Intel® QATzip comes with some tuning example conf files to use. you can replace the old conf file(under /etc/) by them. The detailed info about Configurable options, please refer Programmer's Guide manual.

The process section name(in configuration file) is the key change for QATzip. There are two way to change:

• QAT Driver default conf file does not contain a [SHIM] section which the Intel® QATzip requires by default. You can follow below step to replace them.
• The default section name in the QATzip can be modified if required by setting the environment variable "QAT_SECTION_NAME".

To update the configuration file, copy the configure file(s) from the directory of $QZ_ROOT/config_file/$YOUR_PLATFORM/$CONFIG_TYPE/*.conf to the directory of /etc

YOUR_PLATFORM: the QAT hardware platform, c6xx for Intel® C62X Series Chipset, dh895xcc for Intel® Communications Chipset 8925 to 8955 Series

CONFIG_TYPE: tuned configure file(s) for different usage, multiple_process_opt for multiple process optimization, multiple_thread_opt for multiple thread optimization.

Restart QAT driver

    service qat_service restart

With current configuration, each PCI-e device in C6XX platform could support 32 processes in maximum.

Enable qzstd

If you want to enable lz4s + postprocessing pipeline, you have to compile qzstd which is a sample app to support ZSTD format compression/decompression. Before enabling qzstd, make sure that you have installed zstd static lib.

Compile qzstd

    cd $QZ_ROOT
    ./autogen.sh
    ./configure --enable-lz4s-postprocessing
    make clean
    make qzstd

test qzstd

    qzstd $your_input_file

Test QATzip

Run the following command to check if the QATzip is setup correctly for compressing or decompressing files:

    qzip -k $your_input_file  -O gzipext -A deflate

File compression in 7z:

    qzip -O 7z FILE1 FILE2 FILE3... -o result.7z

Dir compression in 7z:

    qzip -O 7z DIR1 DIR2 DIR3... -o result.7z

Decompression file in 7z:

    qzip -d result.7z

Dir Decompression with -R:

If the DIR contains files that are compressed by qzip and using gzip/gzipext format, then it should be add -R option to decompress them:

    qzip -d -R DIR

Performance Test With QATzip

Please run the QATzip (de)compression performance test with the following command. Please update the drive configuration and process/thread argument in run_perf_test.sh before running the performance test. Note that when number for threads changed, the argument "max_huge_pages_per_process" in run_perf_test.sh should be changed accordingly, at least 6 times of threads number.

    cd $QZ_ROOT/test/performance_tests
    ./run_perf_test.sh

QATzip API Manual

Please refer to file QATzip-man.pdf under the docs folder Please refer to the link for QAT documents

Open Issues

Known issues relating to the QATzip are described in this section.

QATAPP-26069

Title	Buffers allocated with qzMalloc() can't be freed after calling qzMemDestory
Reference	QATAPP-26069
Description	If the users call qzFree after qzMemDestory, they may encounter free memory error "free(): invalid pointe"
Implication	User use qzMalloc API to allocate continuous memory
Resolution	Ensure qzMemDestory is invoked after qzFree, now we use attribute destructor to invoke qzMemDestory
Affected OS	Linux

Intended Audience

The target audience is software developers, test and validation engineers, system integrators, end users and consumers for QATzip integrated Intel® Quick Assist Technology

Legal

Intel® disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel® representative to obtain the latest forecast , schedule, specifications and roadmaps.

The products and services described may contain defects or errors known as errata which may cause deviations from published specifications. Current characterized errata are available on request.

Copies of documents which have an order number and are referenced in this document may be obtained by calling 1-800-548-4725 or by visiting www.intel.com/design/literature.htm.

Intel, the Intel logo are trademarks of Intel Corporation in the U.S. and/or other countries.

*Other names and brands may be claimed as the property of others