lance-format / lance-format/lance

Lance IO performance is 20x worse than TFRecord

Open
#4,381 4 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

performance
Dominant language
Rust
Stars
7.1k
Forks
852
Avg merge
3d 18h
Merged PRs (30d)
272

Description

I plan to train a model using the TensorFlow training framework, using either Lance or TFRecord for data. I ran a test and found that Lance's pure I/O performance is significantly worse than TFRecord's.
TFRecord's performance is approximately 20 times faster that of Lance.
Below is the results and attached is the test code.

Image Image

I don't know Lance very well, is this as expected?

dataloader_benchmark.py.txt

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the attached dataloader_benchmark.py.txt and reproduce the reported Lance-versus-TFRecord I/O comparison in the TensorFlow training context. Check the benchmark inputs and measurements before tracing the relevant Lance I/O path; done means explaining whether the gap is expected or identifying a reproducible performance problem with evidence.

Written by the indexing model from the issue text.

Assessment

Tech stack
machine-learning, python, tensorflow
Domain
data-engineering, machine-learning, performance
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.