Replies: 9 comments
-
I am not sure we are talking about the same software package. |
Beta Was this translation helpful? Give feedback.
-
Yes, I am writing indexes on disk. |
Beta Was this translation helpful? Give feedback.
-
Gentle reminder |
Beta Was this translation helpful? Give feedback.
-
Gentle Reminder#2 |
Beta Was this translation helpful? Give feedback.
-
I have implemented a simple replication technique in order to implement HA/Distributed Architecture for matching vectors. |
Beta Was this translation helpful? Give feedback.
-
I would love to hear some of your thoughts. |
Beta Was this translation helpful? Give feedback.
-
Implementing a distributed layer on top of Faiss is not something that we, the Faiss library authors, can help you with, sorry. You might want to look into already existing third party libraries which may do something similar, such as Milvus. |
Beta Was this translation helpful? Give feedback.
-
The question is about strategy and not asking you to implement. What do you think would be the best strategy since you guys know about faiss from deep down? |
Beta Was this translation helpful? Give feedback.
-
Milvus and OpenSearch are some examples of distributed Faiss. |
Beta Was this translation helpful? Give feedback.
-
I am aware that FAISS is a file system database and returns an id for every record inserted. My question is related to the best practice/strategy for implementing HA and make it distributed.
My strategy:
All insertion requests are duplicated on other nodes.
In case node A is down. Next node is selected as master i.e node B and insertion requests are duplicated on other nodes.
This is all in theory and before writing code to handle this scenario I am thinking can all this be made simpler?
Is this too complex? Can this be made better?
Thanks.
Beta Was this translation helpful? Give feedback.
All reactions