|
I have my own OSM tile server that I set up following the instructions (with local changes) on switch2osm here: https://switch2osm.org/serving-tiles/manually-building-a-tile-server-14-04/ It's hosted on "the cloud", including using an iSCSI drive for the data. Recently, a network error caused the disk to crash. Luckily, we could save it, but the server was down for a day. Now I am looking to improve the situation to provide better service in 2 respects:
As you can imagine, the performance for creating tiles is not great. Sometimes it takes HOURS. We also want to start using the tile server in production and cannot reasonably allow days of downtime. Are there any reference setups, best practices, etc.? Anyone want to share their hardware configuration? We have thought of a few options:
Any and all opinions, recommendations, and details would be welcome! Thanks, Harold Ship |
|
Local SSD is a must. You haven't said if you're running a world-wide server or just a regional one. Precomputing all tiles world-wide is not feasible. You'd usually pre-compute z0-12 or 0-13 and then render the rest on-demand. Having multiple identical servers helps against server failure, but you might also get away with having a "clone" of the main server that keeps only the rendering database and not the update tables, so if the main server fails the clone can at least produce tiles, just not update them. 1
And iSCSI is just a terrible idea for both performance and reliability. Do not use iSCSI unless you can be very lax on both those criterias, or unless you're using software that is specifically tailored for this kind of hardware (Postgresql isn't). I ignored that warning once and we paid that with cripling downtimes and limiting performance. We'll be glad when those servers get decomissioned this year.
(24 Jul '15, 10:50)
Vincent de P... ♦
Ok I will push for the SSD. When you say it's a must, do you mean for the DB or also for the tiles cache?
(26 Jul '15, 05:28)
haroldship1
Yes, we are trying to run the whole world, so precomputing up to say level 13 might be what we'll do. I guess if/once we have SSD we can re-evaluate the performance.
(26 Jul '15, 05:29)
haroldship1
Well, what happened for us was a network switch died and corrupted our iSCSI volume. So I guess I am also learning the hard way. Luckily fsck recovered it.
(26 Jul '15, 05:31)
haroldship1
|
|
Well, this doesn't seem to be an FAQ about OSM, but a need for server experts. Have a look at Commercial_OSM_Software_and_Services for instant advices. After own internet seraching I only found: http://www.geofabrik.de/media/2012-09-08-osm2pgsql-performance.pdf Some server hardware stuff is mentioned there. Asking the guys from geofabrik directly about your aim cannot be bad. Thank you I'll use this to negotiate with the IT department.
(26 Jul '15, 05:32)
haroldship1
|
