Update TileDB benchmarks

[jparismorgan]

What

Updates the TileDB benchmarks.

Results

How to run yourself

• Launch Amazon Linux x86 r6i.16xlarge EC2 instance.
• SSH in
• Install things:

sudo yum update -y

sudo yum install git -y

(maybe - check if you already have this) sudo yum install python3.11 -y
sudo yum install python3.11-pip -y
If you get mismatched Python versions
	sudo ln -sf /usr/bin/python3.11 /usr/bin/python
	sudo ln -sf /usr/bin/pip3.11 /usr/bin/pip
	sudo ln -sf /usr/bin/python3.11 /usr/bin/python3
	sudo ln -sf /usr/bin/pip3.11 /usr/bin/pip3

(maybe - check if you already have this) sudo yum install docker -y
sudo service docker start
sudo usermod -a -G docker ec2-user
(Log back out and log back in)
(Run this and make sure you see "docker") groups

mkdir repo
cd repo
git clone https://github.com/TileDB-Inc/ann-benchmarks.git
cd ann-benchmarks
git checkout npapa/tiledb
pip3 install -r requirements.txt

Then to run TileDB:

• python install.py --algorithm tiledb
• pip3 install requests==2.28.1
  • Fix for: urllib3 v2 incompatibility docker/docker-py#3113 (comment)
• python run.py --dataset sift-128-euclidean --algorithm tiledb-ivf-flat --force --batch
  • Note that only --batch mode works well currently, we need to investigate query mode. It will run but our performance is quite bad.
• sudo chmod -R 777 results/sift-128-euclidean/10/tiledb-ivf-flat
  • Fix for permissions error on in results/sift-128-euclidean/10/tiledb-ivf-flat
• python create_website.py
• Download sift-128-euclidean_10_euclidean-batch.html and sift-128-euclidean_10_euclidean-batch.png

Then to run all algorithms:

• python install.py --algorithm tiledb
• (note it will take a while, you can leave it overnight) python run.py --dataset sift-128-euclidean --batch

[NikolaosPapailiou]

Thanks for updating this!

[NikolaosPapailiou]

We can update this to use the numpy arrays directly for ingestion

[jparismorgan]

Thanks, added this as a TODO and will fix this the next time I set up an EC2 instance and run this. Doing this b/c it seems nice to get this PR merged versus leaving in flux until then.

[NikolaosPapailiou]

I think we can just use the defaults for most params in ingestion

[jparismorgan]

Thanks, added this as a TODO and will fix this the next time I set up an EC2 instance and run this. Doing this b/c it seems nice to get this PR merged versus leaving in flux until then.

[NikolaosPapailiou]

dtype, memory_budget args could be removed as this is the default behavior.

[jparismorgan]

Thanks, added this as a TODO and will fix this the next time I set up an EC2 instance and run this. Doing this b/c it seems nice to get this PR merged versus leaving in flux until then.