[Tile-serving] [openstreetmap/mod_tile] Time spent per tile changes to 4.2 seconds when there is a queue (Issue #333)

[Paul Norman]

In an attempt to isolate what's going on, I left the load normal and used render_list to render z17 tiles. It started off behaving as above, with 4 seconds per tile rendered and CPU usage <50%. Then without changing anything, throughput spiked up, time/tile went down, and CPU usage went to 100%.

![image](https://github.com/openstreetmap/mod_tile/assets/1190866/a278a4cb-0f60-4828-a30b-126029eac01a)

render_list was with `-n 80` and num_threads is 120, and the machine has 80 hardware threads

In an effort to reduce the problem down, I shut off apache2/mod_tile and tried render_list with an assortment of -n. Increasing -n doesn't increase throughput or CPU usage, except that sometimes it goes from not running at capacity to generating more tiles and being CPU bottlenecked.

I'm wondering, could some kind of lock be limiting how many requests per second it can deal with? But that wouldn't explain the bursts of good performance.

-- 
Reply to this email directly or view it on GitHub:
https://github.com/openstreetmap/mod_tile/issues/333#issuecomment-1720312015
You are receiving this because you are subscribed to this thread.

Message ID: <openstreetmap/mod_tile/issues/333/1720312015 at github.com>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tile-serving/attachments/20230914/4273e7ee/attachment.htm>