Very Slow OSM Tile Rendering

Question

I have built my own OSM server of North America using switch2osm's Ubuntu 20.04 guide and all went well. Except when it comes to rendering tiles...or pre-rendering.

Here is the most recent completed rendering of a tile: renderd[1019003]: DEBUG: DONE TILE default 10 312-319 336-343 in 1886.877 seconds

Specs: Bare metal - Intel(R) Xeon(R) CPU E5-2630 0 @ 2.30GHz (24 cores) 32GB RAM 1TB SATA sitting on our internal 10gig network

After some googling, I found this git page to help create indexes: https://raw.githubusercontent.com/gravitystorm/openstreetmap-carto/master/indexes.sql

I also used the OSM Wiki page to tune Postgres to the following:

max_connections = 100
shared_buffers = 12GB
work_mem = 4GB
maintenance_work_mem = 10GB
autovacuum_work_mem = 1GB
checkpoint_completion_target = 0.95
effective_cache_size = 8GB

Below, using top -i here is what I am seeing.

I have reached a point where I am not sure what I am doing wrong or what else I can do to aid in making it faster.

Answer 1

Specs: Bare metal - Intel(R) Xeon(R) CPU E5-2630 0 @ 2.30GHz (24 cores) 32GB RAM 1TB SATA

You don't specify what type of drive "1TB SATA" means. Is that a 1TB SATA SSD or HDD? If HDD, then you should definitely switch to SSD, preferably even NVMe. Many file or database operations in the rendering process require fast random access to disk, something an HDD drive won't be able to deliver. This will compound when you run out of RAM, which is quite likely with 32GB and a large OpenStreetMap extract, as your system may start to swap as well.

HDD is only suitable for backups nowadays, not to process something the size of a North America extract of OpenStreetMap, and the price of NVMe really shouldn't matter nowadays as well, buying something like a 2TB NVMe drive isn't prohibitively expensive any more.