NVIDIA HierarchicalKV

About HierarchicalKV

HierarchicalKV is a part of NVIDIA Merlin and provides hierarchical key-value storage to meet RecSys requirements.

The key capability of HierarchicalKV is to store key-value (feature-embedding) on high-bandwidth memory (HBM) of GPUs and in host memory.

You can also use the library for generic key-value storage.

Benefits of HierarchicalKV

When building large recommender systems, machine learning (ML) engineers face the following challenges:

• GPUs are needed, but HBM on a single GPU is too small for the large DLRMs that scale to several terabytes.
• Improving communication performance is getting more difficult in larger and larger CPU clusters.
• It is difficult to efficiently control consumption growth of limited HBM with customized strategies.
• Most generic key-value libraries provide low HBM and host memory utilization.

HierarchicalKV alleviates these challenges and helps the machine learning engineers in RecSys with the following benefits:

• Supports training large RecSys models on HBM and host memory at the same time.
• Provides better performance by full bypassing CPUs and reducing the communication workload.
• Implements table-size restraint strategies that are based on LRU or customized strategies. The strategies are implemented by CUDA kernels.
• Operates at a high working-status load factor that is close to 1.0.

HierarchicalKV makes NVIDIA GPUs more suitable for training large and super-large models of search, recommendations, and advertising. The library simplifies the common challenges to building, evaluating, and serving sophisticated recommenders models.

API Documentation

The main classes and structs are below, and it's recommended to read the comments in the source code directly:

• class HashTable
• class EvictStrategy
• struct HashTableOptions
• Struct HashTable::Vector

For regular API doc, please refer to API Docs

Usage restrictions

• The key_type and meta_type must be uint64_t.

Contributors

HierarchicalKV is co-maintianed by NVIDIA Merlin Team and NVIDIA product end-users, and also open for public contributions, bug fixes, and documentation. [Contribute]

How to build

Basically, HierarchicalKV is a headers only library, the commands below only create binaries for benchmark and unit testing.

git clone --recursive https://github.com/NVIDIA-Merlin/HierarchicalKV.git
cd HierarchicalKV && mkdir -p build && cd build
cmake -DCMAKE_BUILD_TYPE=Release -Dsm=80 .. && make -j

// for Benchmark
./merlin_hashtable_benchmark

// for Unit Test
./merlin_hashtable_test

Your environment must meet the following requirements:

• CUDA version >= 11.2
• NVIDIA GPU with compute capability 8.0, 8.6, 8.7 or 9.0

Benchmark

• GPU: 1 x NVIDIA A100-SXM4-80GB: 8.0
• Key Type = uint64_t
• Value Type = float32 * {dim}
• Key-Values per OP = 1,048,576
• Throughput Unit: Billion-KV/second

On pure HBM mode:

dim	capacity	load_factor	HBM(GB)	HMEM(GB)	insert	find
4	67108864	0.50	16	0	1.167	1.750
4	67108864	0.75	16	0	0.897	1.386
4	67108864	1.00	16	0	0.213	0.678
16	67108864	0.50	16	0	1.114	1.564
16	67108864	0.75	16	0	0.894	1.258
16	67108864	1.00	16	0	0.215	0.640
64	67108864	0.50	16	0	0.873	0.915
64	67108864	0.75	16	0	0.767	0.823
64	67108864	1.00	16	0	0.206	0.492
128	134217728	0.50	64	0	0.664	0.613
128	134217728	0.75	64	0	0.593	0.560
128	134217728	1.00	64	0	0.191	0.387

On HBM+HMEM hybrid mode:

dim	capacity	load_factor	HBM(GB)	HMEM(GB)	insert	find
64	134217728	0.50	16	16	0.107	0.103
64	134217728	0.75	16	16	0.106	0.101
64	134217728	1.00	16	16	0.077	0.094
64	1073741824	0.50	56	200	0.037	0.040
64	1073741824	0.75	56	200	0.037	0.040
64	1073741824	1.00	56	200	0.030	0.036
128	67108864	0.50	16	16	0.076	0.072
128	67108864	0.75	16	16	0.071	0.071
128	67108864	1.00	16	16	0.059	0.068
128	536870912	0.50	56	200	0.039	0.040
128	536870912	0.75	56	200	0.041	0.040
128	536870912	1.00	56	200	0.035	0.038

Support and Feedback:

If you encounter any issues or have questions, go to https://github.com/NVIDIA-Merlin/HierarchicalKV/issues and submit an issue so that we can provide you with the necessary resolutions and answers.

Acknowledgment

We are very grateful to external initial contributors @Zhangyafei and @Lifan for their design, coding, and review work.

License

Apache License 2.0