k-NN index

The k-NN plugin introduces a custom data type, the knn_vector, that allows users to ingest their k-NN vectors into an OpenSearch index and perform different kinds of k-NN search. The knn_vector field is highly configurable and can serve many different k-NN workloads. For more information, see k-NN vector.

Lucene byte vector

Starting with k-NN plugin version 2.9, you can use byte vectors with the lucene engine in order to reduce the amount of storage space needed. For more information, see Lucene byte vector.

Method definitions

A method definition refers to the underlying configuration of the Approximate k-NN algorithm you want to use. Method definitions are used to either create a knn_vector field (when the method does not require training) or create a model during training that can then be used to create a knn_vector field.

A method definition will always contain the name of the method, the space_type the method is built for, the engine (the library) to use, and a map of parameters.

Mapping parameterRequiredDefaultUpdatableDescription
nametruen/afalseThe identifier for the nearest neighbor method.
space_typefalsel2falseThe vector space used to calculate the distance between vectors.
enginefalsenmslibfalseThe approximate k-NN library to use for indexing and search. The available libraries are faiss, nmslib, and Lucene.
parametersfalsenullfalseThe parameters used for the nearest neighbor method.

Supported nmslib methods

Method nameRequires trainingSupported spacesDescription
hnswfalsel2, innerproduct, cosinesimil, l1, linfHierarchical proximity graph approach to Approximate k-NN search. For more details on the algorithm, see this abstract.

HNSW parameters

Parameter nameRequiredDefaultUpdatableDescription
ef_constructionfalse512falseThe size of the dynamic list used during k-NN graph creation. Higher values lead to a more accurate graph but slower indexing speed.
mfalse16falseThe number of bidirectional links that the plugin creates for each new element. Increasing and decreasing this value can have a large impact on memory consumption. Keep this value between 2 and 100.

For nmslib, ef_search is set in the index settings.

Supported faiss methods

Method nameRequires trainingSupported spacesDescription
hnswfalsel2, innerproductHierarchical proximity graph approach to Approximate k-NN search.
ivftruel2, innerproductBucketing approach where vectors are assigned different buckets based on clustering and, during search, only a subset of the buckets is searched.

For hnsw, “innerproduct” is not available when PQ is used.

HNSW parameters

Parameter nameRequiredDefaultUpdatableDescription
ef_searchfalse512falseThe size of the dynamic list used during k-NN searches. Higher values lead to more accurate but slower searches.
ef_constructionfalse512falseThe size of the dynamic list used during k-NN graph creation. Higher values lead to a more accurate graph but slower indexing speed.
mfalse16falseThe number of bidirectional links that the plugin creates for each new element. Increasing and decreasing this value can have a large impact on memory consumption. Keep this value between 2 and 100.
encoderfalseflatfalseEncoder definition for encoding vectors. Encoders can reduce the memory footprint of your index, at the expense of search accuracy.

IVF parameters

Parameter nameRequiredDefaultUpdatableDescription
nlistfalse4falseNumber of buckets to partition vectors into. Higher values may lead to more accurate searches at the expense of memory and training latency. For more information about choosing the right value, refer to Guidelines to choose an index.
nprobesfalse1falseNumber of buckets to search during query. Higher values lead to more accurate but slower searches.
encoderfalseflatfalseEncoder definition for encoding vectors. Encoders can reduce the memory footprint of your index, at the expense of search accuracy.

For more information about setting these parameters, refer to the Faiss documentation.

IVF training requirements

The IVF algorithm requires a training step. To create an index that uses IVF, you need to train a model with the Train API, passing the IVF method definition. IVF requires that, at a minimum, there should be nlist training data points, but it is recommended that you use more. Training data can be composed of either the same data that is going to be ingested or a separate dataset.

Supported Lucene methods

Method nameRequires trainingSupported spacesDescription
hnswfalsel2, cosinesimilHierarchical proximity graph approach to Approximate k-NN search.

HNSW parameters

Parameter nameRequiredDefaultUpdatableDescription
ef_constructionfalse512falseThe size of the dynamic list used during k-NN graph creation. Higher values lead to a more accurate graph but slower indexing speed.
The Lucene engine uses the proprietary term “beam_width” to describe this function, which corresponds directly to “ef_construction”. To be consistent throughout OpenSearch documentation, we retain the term “ef_construction” to label this parameter.
mfalse16falseThe number of bidirectional links that the plugin creates for each new element. Increasing and decreasing this value can have a large impact on memory consumption. Keep this value between 2 and 100.
The Lucene engine uses the proprietary term “max_connections” to describe this function, which corresponds directly to “m”. To be consistent throughout OpenSearch documentation, we retain the term “m” to label this parameter.

Lucene HNSW implementation ignores ef_search and dynamically sets it to the value of “k” in the search request. Therefore, there is no need to make settings for ef_search when using the Lucene engine.

  1. {
  2. "type": "knn_vector",
  3. "dimension": 100,
  4. "method": {
  5. "name":"hnsw",
  6. "engine":"lucene",
  7. "space_type": "l2",
  8. "parameters":{
  9. "m":2048,
  10. "ef_construction": 245
  11. }
  12. }
  13. }

Supported faiss encoders

You can use encoders to reduce the memory footprint of a k-NN index at the expense of search accuracy. faiss has several encoder types, but the plugin currently only supports flat and pq encoding.

The following example method definition specifies the hnsw method and a pq encoder:

  1. "method": {
  2. "name":"hnsw",
  3. "engine":"faiss",
  4. "space_type": "l2",
  5. "parameters":{
  6. "encoder":{
  7. "name":"pq",
  8. "parameters":{
  9. "code_size": 8,
  10. "m": 8
  11. }
  12. }
  13. }
  14. }

The hnsw method supports the pq encoder for OpenSearch versions 2.10 and later. The code_size parameter of a pq encoder with the hnsw method must be 8.

Encoder nameRequires trainingDescription
flatfalseEncode vectors as floating point arrays. This encoding does not reduce memory footprint.
pqtrueAn abbreviation for product quantization, it is a lossy compression technique that uses clustering to encode a vector into a fixed size of bytes, with the goal of minimizing the drop in k-NN search accuracy. At a high level, vectors are broken up into m subvectors, and then each subvector is represented by a code_size code obtained from a code book produced during training. For more information about product quantization, see this blog post.

Examples

The following example uses the ivf method without specifying an encoder (by default, OpenSearch uses the flat encoder):

  1. "method": {
  2. "name":"ivf",
  3. "engine":"faiss",
  4. "space_type": "l2",
  5. "parameters":{
  6. "nlist": 4,
  7. "nprobes": 2
  8. }
  9. }

The following example uses the ivf method with a pq encoder:

  1. "method": {
  2. "name":"ivf",
  3. "engine":"faiss",
  4. "space_type": "l2",
  5. "parameters":{
  6. "encoder":{
  7. "name":"pq",
  8. "parameters":{
  9. "code_size": 8,
  10. "m": 8
  11. }
  12. }
  13. }
  14. }

The following example uses the hnsw method without specifying an encoder (by default, OpenSearch uses the flat encoder):

  1. "method": {
  2. "name":"hnsw",
  3. "engine":"faiss",
  4. "space_type": "l2",
  5. "parameters":{
  6. "ef_construction": 256,
  7. "m": 8
  8. }
  9. }

PQ parameters

Paramater NameRequiredDefaultUpdatableDescription
mfalse1falseDetermine how many many sub-vectors to break the vector into. sub-vectors are encoded independently of each other. This dimension of the vector must be divisible by m. Max value is 1024.
code_sizefalse8falseDetermines the number of bits to encode a sub-vector into. Max value is 8. Note — for IVF, this value must be less than or equal to 8. For HNSW, this value can only be 8.

Choosing the right method

There are a lot of options to choose from when building your knn_vector field. To determine the correct methods and parameters to choose, you should first understand what requirements you have for your workload and what trade-offs you are willing to make. Factors to consider are (1) query latency, (2) query quality, (3) memory limits, (4) indexing latency.

If memory is not a concern, HNSW offers a very strong query latency/query quality tradeoff.

If you want to use less memory and index faster than HNSW, while maintaining similar query quality, you should evaluate IVF.

If memory is a concern, consider adding a PQ encoder to your HNSW or IVF index. Because PQ is a lossy encoding, query quality will drop.

Memory estimation

In a typical OpenSearch cluster, a certain portion of RAM is set aside for the JVM heap. The k-NN plugin allocates native library indexes to a portion of the remaining RAM. This portion’s size is determined by the circuit_breaker_limit cluster setting. By default, the limit is set at 50%.

Having a replica doubles the total number of vectors.

HNSW memory estimation

The memory required for HNSW is estimated to be 1.1 * (4 * dimension + 8 * M) bytes/vector.

As an example, assume you have a million vectors with a dimension of 256 and M of 16. The memory requirement can be estimated as follows:

  1. 1.1 * (4 * 256 + 8 * 16) * 1,000,000 ~= 1.267 GB

IVF memory estimation

The memory required for IVF is estimated to be 1.1 * (((4 * dimension) * num_vectors) + (4 * nlist * d)) bytes.

As an example, assume you have a million vectors with a dimension of 256 and nlist of 128. The memory requirement can be estimated as follows:

  1. 1.1 * (((4 * 256) * 1,000,000) + (4 * 128 * 256)) ~= 1.126 GB

Index settings

Additionally, the k-NN plugin introduces several index settings that can be used to configure the k-NN structure as well.

At the moment, several parameters defined in the settings are in the deprecation process. Those parameters should be set in the mapping instead of the index settings. Parameters set in the mapping will override the parameters set in the index settings. Setting the parameters in the mapping allows an index to have multiple knn_vector fields with different parameters.

SettingDefaultUpdatableDescription
index.knnfalsefalseWhether the index should build native library indexes for the knn_vector fields. If set to false, the knn_vector fields will be stored in doc values, but Approximate k-NN search functionality will be disabled.
index.knn.algo_param.ef_search512trueThe size of the dynamic list used during k-NN searches. Higher values lead to more accurate but slower searches. Only available for nmslib.
index.knn.algo_param.ef_construction512falseDeprecated in 1.0.0. Use the mapping parameters to set this value instead.
index.knn.algo_param.m16falseDeprecated in 1.0.0. Use the mapping parameters to set this value instead.
index.knn.space_typel2falseDeprecated in 1.0.0. Use the mapping parameters to set this value instead.