Greg Hanson

> I'm not sure why but parallelization has actually made it slower. Maybe it makes caching harder or it has to move too much data in- and out of threads....

> Also not sure how serde performs deserializing larger files in comparison to to any other crates so might be room for improvement there as well. For the purposes of...

Sorry I didn't mention it from the get-go, but for testing I have been using an AWS EC2 instance, specifically a [x2iedn.2xlarge](https://aws.amazon.com/ec2/instance-types/x2i/) with EBS storage | vCPU | Mem (GiB)...

> I wonder what the C++ is exactly. Taking a closer look at their implementation is on my TODO list and may be the next thing I look into. I...

@Tpt sorry for the delay on this one - I am curious to hear if this implementation matches with what you envisioned. > the HDT implementation can just return the...

cc @donpellegrino as this is iterative of his original implementation

Thanks for the advice! that does simplify things quite a bit > This way your implementation does not depend on the oxigraph crate at all and you can use a...

> Ideally, the approach would still allow the use of HDT as read-only backend with equivalent architectural standing to other backends, such as RocksDB. With that approach, other Oxigraph capabilities...

Sorry for the lack of updates here. I haven't been too impressed with responsiveness and release cadence of the [hdt-cpp](https://github.com/rdfhdt/hdt-cpp) library, so I just decided to re-write it in Rust....

## binaries I think separate binaries would be the way to go, you can define multiple `[[bin]]` sections in the Cargo.toml for each one. I say this because it operates...