Tuning Guide for RocksDB Compression and Decompression with Intel® IAA and 4th Generation Intel® Xeon® Scalable Processors




This guide is targeted towards users who are already familiar with RocksDB and provides pointers and system setting for hardware and software that will provide the best performance for most situations. However, please note that we rely on the users to carefully consider these settings for their specific scenarios since RocksDB can be deployed in multiple ways and this is a reference to one such use case.

The 4th generation Intel® Xeon® Scalable processors deliver workload-optimized performance with built-in acceleration for AI, encryption, HPC, storage, database systems, and networking. They feature unique security technologies to help protect data on premises or in the cloud. Improvements of particular interest to this workload include:

  • New built-in accelerators for AI, HPC, networking, security, storage, and analytics.
  • Support for Intel® In-Memory Analytics Accelerator (Intel® IAA)
  • New flex bus input/output (I/O) interface (PCIe* 5.0 and CXL)
  • New flexible I/O interface up to 20 HSIO lanes (PCIe 3.0)
  • Increased I/O bandwidth with PCIe 5.0 (up to 80 lanes)
  • Increased memory bandwidth with DDR5
  • Increased multisocket bandwidth with UPI 2.0 (up to 16 GT/s)

Server Configuration


This configuration is based on the 4th generation Intel Xeon processor. The server platform, memory, hard drives, and network interface cards can be determined according to your usage requirements.




4th generation Intel Xeon Scalable processors, base frequency 1.6 GHz




1024 GB (16x64 GB, 4800 MT/s [4800 MT/s])

Storage & Disks

Intel® Optane™ SSD DC P5800X series





Operating System

CentOS* Stream 8


5.18 or later

GNU Compiler Collection (GCC)* Version

gcc-8.5.0 2021-05-14 (Red Hat* 8.5.0-15)

GNU C Library (glibc) Version

ldd 2.28

GNU Binutils Version

GNU ld version 2.36.1-3.el8

Introduction to RocksDB

RocksDB started as an experimental Facebook* project that was intended to develop a database product for high-workload services, which is comparable in data storage performance to a fast data storage system, flash memory in particular. RocksDB uses source code from the open source project, LevelDB, as well as foundational concepts from the Apache HBase* project. The source code can be sourced back to the LevelDB 1.5 fork.

RocksDB supports various configurations and can be tuned for different production environments (memory only, flash drives, hard disks, or Hadoop Distributed File System [HDFS]). It also supports a range of data compression algorithms and tools for debugging production environments. RocksDB is mainly designed for excellent performance in fast storage applications and high workload services. So, this database system requires sufficient use of the read and write speeds of flash memory and RAM. RocksDB needs to support efficient point-lookup and range-scan operations, and configuration of various parameters for random read and write operations in high-workload services or for performance tuning in case of heavy read and write traffic.

For more information on RocksDB, see RocksDB wiki.

Introduction to Intel IAA 

This analytics accelerator is focused on data analytics operations (decompress, compress, and filter).

  • 2x–4x usable memory capacity
  • Lower CPU use ((de)compression, and analytics and SQL offload)
  • Lower power consumption
  • Use cases
    • Memory caching (memcached and Redis*)
    • In memory databases and analytics (Apache Spark* SQL and Apache Cassandra*)
    • In storage select (Amazon* S3)
    • In SSD analytics, and more

In this RocksDB tuning feature, Intel IAA (de)compression is enabled in RocksDB to optimize data access performance and storage.

Hardware Tuning

This guide targets the usage of RocksDB on 4th generation Intel Xeon Scalable processors with Intel IAA.

IAA 1.0 Technical Specification was published in February 2022. The specification passed all technical and BU approvals. It is being processed for further publication on Software Developer Manuals.

Intel IAA Specification

BIOS Setting

Begin by resetting your BIOS to default setting, then change the default values to the below suggestions:

Configuration Item

Recommended Value



Hardware Prefetcher


L2 RFO Prefetch Disable


Adjacent Cache Prefetch


DCU Streamer Prefetcher


DCU IP Prefetcher


LLC Prefetch


Total Memory Encryption (TME)




UMA-Based Clustering

Hemisphere (2-clusters)

Boot performance mode

Max Performance

Turbo Mode


Hardware P-State

Native Mode

Local/Remote Threshold



Intel IAA Settings

accel-config is a user-space tool for controlling and configuring Intel IAA hardware devices.

To install the accel-config tool, see idxd-config.

List all the active DSA devices discovered by the driver and the driver exported attributes.

accel-config list

The following is the script to enable one Intel IAA device, configure Intel IAA work queues, and engines resources:

#Enable 1 IAA devices
for ((i=1;$i<=1;i=$i+2)) # modify this value to enable different number of IAA devices
    echo $i
    accel-config config-engine iax$i/engine$i.0 -g 0
    accel-config config-engine iax$i/engine$i.1 -g 0
    accel-config config-engine iax$i/engine$i.2 -g 0
    accel-config config-engine iax$i/engine$i.3 -g 0
    accel-config config-engine iax$i/engine$i.4 -g 0
    accel-config config-engine iax$i/engine$i.5 -g 0
    accel-config config-engine iax$i/engine$i.6 -g 0
    accel-config config-engine iax$i/engine$i.7 -g 0
    accel-config config-wq iax$i/wq$i.0 -g 0 -s 128 -p 10 -b 1 -d user -t 128 -m shared -y user -n iax_crypto
    accel-config enable-device iax$i
    accel-config enable-wq iax$i/wq$i.0

The following is an Intel IAA device configuration output:


Network Configuration

In the application scenario of the RocksDB, since performance is usually limited by the bandwidth of the network rather than the performance of memory and persistent memory, running Redis across the network requires high network card bandwidth (the larger, the better), which is recommended to be higher than 10 GB/s.

Software Tuning

Software configuration tuning is essential. From the operating system to the RocksDB configuration, software settings are designed for general-purpose applications. These default settings are almost never tuned for the best performance.

Linux* Kernel Optimization Settings

CPU Configuration

  • Configure the CPU to the performance mode:
cpupower -c <cpulist> frequency-set --governor performance
  • Configure the energy and performance bias:
x86_energy_perf_policy performance
  • Configure the minimum value for the processor P-State:
echo 100 > /sys/device/system/cpu/intel_pstate/min_perf_pct

Kernel Settings

  • Configure the corresponding CPU core parameters:
sysctl -w kernel.sched_domain.cpu<x>.domain0.max_newidle_lb_cost=0

sysctl -w kernel.sched_domain.cpu<x>.domain1.max_newidle_lb_cost=0
  • Configure the scheduling granularity:
sysctl -w kernel.sched_min_granularity_ns=10000000

sysctl -w kernel.sched_wakeup_granularity_ns=15000000
  • Configure the virtual memory parameters:
sysctl -w vm.dirty_ratio = 40

sysctl -w vm.swappiness = 10

sysctl -w vm.dirty_background_ratio=10


RocksDB Architecture

RocksDB is a storage engine library of key-value store interfaces where keys and values are arbitrary byte streams and implemented based on the log-structured merge -tree (LSM) algorithm.

RocksDB writes process: Data is added to Write-Ahead Log (WAL) first for every insert or update, then it is inserted into a Memtable for sorting. If the Memtable is already full, it will be converted to an immutable Memtable, and the data is refreshed as the SST file to level 0 in the back end. Similarly, when one level is full, “compaction” is triggered in the back end, which takes the data in the SST file and merges it with a higher-level SST file with an overlapping key range. The merged data takes the new SST file and writes it to a higher level, while the expired data is discarded. Since the higher level is 10 times larger than the lower level, compaction will cause serious write amplification and occupy a large amount of storage bandwidth. This is the main performance issue with LSM trees.

A Read operation starts with a Memtable search. Then a level-by-level search is performed in the SST files until the block data is found. If the block data is already in the block cache, the data is read directly from the cache (cache hit). Otherwise, the data is loaded onto the block cache from the SST file and read (cache miss). Block data is the smallest unit of I/O for a read operation and is usually larger than a key-value pair. Therefore, there will be a certain degree of reading amplification.

RocksDB Tuning

Install Intel® oneAPI Threading Building Blocks (oneTBB)

# Download offline installer from https://www.intel.com/content/www/cn/zh/developer/articles/tool/oneapi-standalone-components.html#onetbb
# If there are other oneAPI products installed in another directory, --install-dir selection will fail
if [[ $install_tbb == "true" ]]; then
  ./$tbb_file -a -s --eula accept --install-dir ./oneapi
  export LD_LIBRARY_PATH=$(pwd)/oneapi/tbb/latest/lib/intel64/gcc4.8
  export LD_LIBRARY_PATH=$tbb_install_dir/lib/intel64/gcc4.8


Install Intel Query Processing Library (Intel QPL)


if [[ $use_git == "true" ]]; then

  git clone --recursive --branch v0.1.21 https://github.com/intel/qpl.git qpl_source


  unzip $qpl_file

  mv $qpl_file_dir qpl_source


cd qpl_source

mkdir build

cd build


cmake --build . --target install -- -j

cd ../..

Install Global Flags Editor (GFlags)

# gflags

# CentOS doesn't offer v2.2, required by latest RocksDB

# Comment out this section if your system includes gflags v2.2 already

if [[ $install_gflags == "true" ]]; then

  git clone https://github.com/gflags/gflags.git

  cd gflags

  git checkout tags/v2.2.2

  mkdir build && cd build



  make install


  cd ../..


Install Other Dependent Libraries

  • Snappy
yum install snappy snappy-devel
  • zlib
yum install zlib zlib-devel
  • bzip2
yum install bzip2 bzip2-devel
  • LZ4
yum install lz4-devel
  • Zstandard
wget https://github.com/facebook/zstd/archive/v1.1.3.tar.gz

mv v1.1.3.tar.gz zstd-1.1.3.tar.gzzxvf zstd-1.1.3.tar.gz

cd zstd-1.1.3

make && make install
  • memkind

memkind v1.10.0 or higher is required to build the block cache allocator. For more information on memkind, see GitHub* memkind.

yum --enablerepo=PowerTools install memkind-devel


Compile RocksDB

Download the RocksDB from Intel, Which Supports Intel IAA Compress

git clone --branch pluggable_compression https://github.com/lucagiac81/rocksdb.git

Download Intel IAA Compressor

git clone https://github.com/intel/iaa-plugin-rocksdb.git plugin/iaa_compressor

Build RocksDB with Intel IAA Compressor

EXTRA_CXXFLAGS="-I/home/user/wangpeng/qpl_19/include -I/home/user/wangpeng/qpl_19/include/qpl -I/usr/local/include" EXTRA_LDFLAGS="-L/home/user/wangpeng/qpl_19/lib64 -L/home/user/wangpeng/oneapi/tbb/latest/lib/intel64/gcc4.8 -L/usr/local/lib -Lgflags" ROCKSDB_CXX_STANDARD="c++17" DISABLE_WARNING_AS_ERROR=1 ROCKSDB_PLUGINS="iaa_compressor" LIB_MODE=shared make -j release


Create a new RocksDB Compression with Intel IAA and Intel QPL

  • RocksDB Compression: RocksDB supports many compressions: LZ4, SNAPPY, ZSTD, and ZLIB. RocksDB uses them to compress a data block.
  • Intel QPL: Intel QPL is aimed to support capabilities of Intel IAA available on 4th generation Intel Xeon Scalable processors, such as very high throughput compression and decompression combined with primitive analytic functions, as well as to provide highly-optimized software fallback on other Intel CPUs.
  • A new compression of Intel IAA and Intel QPL: Create a new RocksDB compressor with the Intel = QPL API, which is used to enable Intel IAA to compress and decompress. This new compressor has the same way to use as other compressions.
  • RocksDB with Intel IAA and Intel QPL Architecture

flow chart shows the structure of Intel IAA and Intel QPL


Read & Write Optimization in RocksDB

flow chart shows the structure of write and read memory and storage

Write Optimization

In RocksDB, the writing operation inserts data to the Memtable first, and then these data is compressed according to the block into an SST file. The compressed performance significantly impacts writing efficiency. Intel IAA has efficient compression and decompression, and it can speed up writing performance. It moves the compression load from the CPU to Intel IAA and reduces CPU use.

Read Optimization

If the reading operation doesn’t search the key from a block cache, it needs to decompress the SST compressed block. Decompression performance significantly impacts reading efficiency. According to the laboratory data, after enabling Intel IAA decompression the reading performance has much improved compared with zstd (CPU) compression.

Compaction Optimization in RocksDB

flow chart shows how RocksDB compacts

The following image is the compaction flowchart. Every compaction process contains compression and decompression which have a significant impact on compactions. After enabling the Intel IAA compress, it can improve compaction performance and reduce CPU use.



The combination of RocksDB, 4th generation Intel Xeon Scalable processor, and Intel IAA is a very practical way to use RocksDB. The large dataset of compression needs a high compress ratio compress method, and Intel IAA supplies this suitable compression way. It can get good performance, while it can be conveniently used in customers’ environment. Intel IAA (de)compression has been created as a pluggable compressor, and it has no code change in the customer’s environment.

If customers use RocksDB and Intel IAA in MySQL or Redis, it can get better performance compared with zstd (CPU) compression, and better storage saving compared with LZ4 (CPU) compression. In the performance experiment, after enabling Intel IAA compression, the compression hot spot significantly drops the data accessing workload compared with other compressions. Meanwhile, RocksDB and Intel IAA can save much of the MySQL* buffer size while keeping the same throughput with zstd (CPU) compression.

Intel IAA compression also has better throughput than zstd (CPU) compression, and it can supply better performance in data writing workloads.

As mentioned in the Facebook for RocksDB introduction, RocksDB provides better data flashing and data compressing in reading, writing, and storing data workloads. Intel IAA (de)compress function can improve these functions.


We value your feedback. If you have comments (positive or negative) on this guide or are seeking something that is not part of this guide, reach out and let us know what you think.