Accelerate image compression¶
A general framework to accommodate hardware compression accelerators for compression of volumes uploaded to the Image service (Glance) as images and decompression of compressed images used to create volumes is introduced in Train release.
The only accelerator supported in this release is Intel QuickAssist Technology (QAT), which produces a compressed file in gzip format. Additionally, the framework provides software-based compression using GUNzip tool if a suitable hardware accelerator is not available. Because this software fallback could cause performance problems if the Cinder services are not deployed on sufficiently powerful nodes, the default setting is not to enable compression on image upload or download.
The compressed image of a volume will be stored in the Image service (Glance)
with the container_format image property of compressed. See the Image
service documentation for more
information about this image container format.
Configure image compression¶
To enable the image compression feature, set the following configuration option
in the cinder.conf file:
allow_compression_on_image_upload = True
By default it will be set to False, which means image compression is disabled.
compression_format = gzip
This is to specify image compression format. The only supported format is
gzip in Train release.
System requirement¶
In order to use this feature, there should be a hardware accelerator existing in system, otherwise no benefit will get from this feature. Regarding the two accelerators that supported, system should be configured as below:
Intel QuickAssist Technology (QAT)- This is the hardware accelerator from Intel. The driver of QAT should be installed, refer to https://01.org/intel-quickassist-technology. Also the compression library QATzip should be installed, refer to https://github.com/intel/QATzip.GUNzip- The related package ofGUNzipshould be installed and the commandgzipshould be available. This is used as fallback when hardware accelerator is not available.
Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.