How can I make Microsoft.SemanticKernel.Memory.SearchAsync faster?

[gusmally]

I have a .sqlite db with 2,500 rows that I'm using as a memory store:

var memoryStore = await SqliteMemoryStore.ConnectAsync(<path to db>);

var memoryBuilder = new MemoryBuilder();
var openAiEmbedder = new OpenAITextEmbeddingGenerationService("text-embedding-3-small",
				_openAiKey);
memoryBuilder.WithTextEmbeddingGeneration(openAiEmbedder);
memoryBuilder.WithMemoryStore(memoryStore);
myItem.Memory = memoryBuilder.Build();

when I later perform myItem.Memory.SearchAsync(collectionName, query, limit: 1000, minRelevanceScore: 0.73);, it takes a couple seconds to run.

Is there a way to speed this up, such as using a different memory store? Have I overlooked anything else?

[dmytrostruk]

@gusmally I would recommend using other memory store, since current implementation for Sqlite fetches all DB records during search operation which is not the best approach in terms of performance:

semantic-kernel/dotnet/src/Connectors/Connectors.Memory.Sqlite/SqliteMemoryStore.cs

Lines 136 to 147 in 305b2ee

	await foreach (var record in this.GetAllAsync(collectionName, cancellationToken).ConfigureAwait(false))
	{
	if (record is not null)
	{
	double similarity = TensorPrimitives.CosineSimilarity(embedding.Span, record.Embedding.Span);
	if (similarity >= minRelevanceScore)
	{
	var entry = withEmbeddings ? record : MemoryRecord.FromMetadata(record.Metadata, ReadOnlyMemory<float>.Empty, record.Key, record.Timestamp);
	embeddings.Add(new(entry, similarity));
	}
	}
	}

Please take a look at other available memory connectors:
https://github.com/microsoft/semantic-kernel/tree/main/dotnet/src/Connectors

We are also working on new version of memory connectors that will allow to use custom schema and are up to date. This work is happening in feature branch at the moment and experimental version of it will be available soon:
https://github.com/microsoft/semantic-kernel/tree/feature-memory-service

[gusmally]

Thank you for the analysis and recommendation!