kevkibe/memvectordb

⚡️Lightning fast in-memory VectorDB written in rust🦀

MemVectorDB

An fast in-memory VectorDB in rust.

⚡️ Features

• Fast: MemVectorDB stores vectors in-memory, ensuring fast insertion and retrieval operations.
• Vertical Scalability: With vectors stored in-memory, MemVectorDB can scale vertically based on available system resources.
• Metadata Support: Supports metadata storage, beneficial for RAG (Retrieval Augmented Generation) applications and pipelines.
• Open Source: MIT Licensed, free forever.

🚀 Usage

1. Clone the repository:

git clone https://github.com/KevKibe/memvectordb.git

2. Build dependencies:

make build

3. Start the server

make run

• Server runs on http://localhost:8000

🐳 Using Docker

1. Pull the Docker image:

• On x86_64 (Intel/AMD) systems:

docker pull kevkibe/memvectordb:v1.5.0

• On ARM-based systems (e.g., M1, M2, M3):

docker pull --platform linux/amd64 kevkibe/memvectordb:v1.5.0

2. Run the Docker container:

• On x86_64 (Intel/AMD) systems:

docker run -p 8000:8000 kevkibe/memvectordb

• On ARM-based systems (e.g., M1, M2, M3):

docker run -p 8000:8000 --platform linux/amd64 kevkibe/memvectordb

3. Server runs on http://localhost:8000

MemVectorDB Python client: Docs

Load Tests

• All tests done with 100000 requests on a Macbook Air M1.

POST

Summary:
  Success rate: 100.00%
  Total:        1.9317 secs
  Slowest:      0.0363 secs
  Fastest:      0.0000 secs
  Average:      0.0010 secs
  Requests/sec: 51766.9796

  Total data:   5.15 MiB
  Size/request: 54 B
  Size/sec:     2.67 MiB

GET

Summary:
  Success rate: 100.00%
  Total:        1.0847 secs
  Slowest:      0.0081 secs
  Fastest:      0.0000 secs
  Average:      0.0005 secs
  Requests/sec: 92191.6443

  Total data:   4.58 MiB
  Size/request: 48 B
  Size/sec:     4.22 MiB

DELETE

Summary:
  Success rate: 100.00%
  Total:        1.0714 secs
  Slowest:      0.0168 secs
  Fastest:      0.0000 secs
  Average:      0.0005 secs
  Requests/sec: 93339.6446

  Total data:   5.05 MiB
  Size/request: 52 B
  Size/sec:     4.72 MiB

PUT

Summary:
  Success rate: 100.00%
  Total:        2.7216 secs
  Slowest:      0.0395 secs
  Fastest:      0.0001 secs
  Average:      0.0014 secs
  Requests/sec: 36743.3094

  Total data:   7.06 MiB
  Size/request: 74 B
  Size/sec:     2.59 MiB

memvectordb's Issues

Enhancement: Option for persistence

Functionality to the system to allow for data persistence, meaning the ability to store data beyond the runtime of the application for user's that require that.

Enhancement: Add logging

• Implement logging of changes made to db instance.

Optimize `CacheDB`.`insert_into_collection` method to O(1) Time Complexity

• replace the linear scan for duplicate IDs with a constant-time operation using HashMap for fast ID lookup.
• Might have to modify the Collection and Embedding structs.
• Goal is to provide O(1) average-time complexity for checking if an embedding ID already exists.

Enhancement: Implement Semantic Search Functionality

This feature is to enhance the functionality of the vector database by introducing semantic search capabilities.
Semantic search enables users to retrieve results that are contextually relevant to their query, beyond just keyword matching.