Geovisio instance's hosting

[antoine-de]

It would be great to have a geovisio instance handled by the team to have a place where to store the first harvested pictures.

Do you have any ideas on what to use for this? It seems that we can have some OVH credits but we are not bound to them.

We need to keep in mind that:

• we'll have quite a lot a data (at least tens of terabytes of it).
• we'll need to run some algorithme to blur the pictures

I see several alternatives:

1. we can keep the compute and the storage in the same place a. using a bare metal b. a vps

2. split the compute and storage a. use a bare metal or vps server and put the pictures on block or object storage b. use a paas (like scalingo to get automated update), and put the pictures in a separate object storage

Do you have other ideas?

We need to check the price for those alternative, and the performance for each of them.

We can also chose something now and change it later we the need arise.

Do you have some ideas on this?

[PanierAvide]

As now, Geovisio instances are mainly deployed on classic servers, with either local filesystem storage for pictures or through local network mounts. Géovélo is working on deploying on a Scaleway hosting with separated API and pictures storage (S3), but first experiments shows that read/write speeds on these S3 storage is making processing a lot slower.

We might be careful on read/write speeds of filesystems we will use, no matter what infrastructure is chosen.

[cquest]

We can also experiment both in parallel: cloud based and bare metal based.

• OSM-FR instance is bare-metal based, with self-managed virtualisation/containerization.
• IGN instance could be cloud-based only.

This allows to compare (perfs, costs, administration), make sure what we create works in both cases, etc.

To start, cloud will clearly be cheaper. On the lon gterm, with a lot of pictures to store, this will have to be reconsidered.