hans/Nominatim
1
0
Fork 1
mirror of https://github.com/osm-search/Nominatim.git synced 2026-02-16 15:47:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

update minimum required RAM to 64GB

Browse Source 
Also adds more background explanation on time and RAM use,
as well as a hint that github issues are not good for
installation support.
This commit is contained in:
Sarah Hoffmann
2019-11-24 10:31:34 +01:00
parent 000fe3ddff
commit 88fab44006
4 changed files with 70 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
docs/admin/Import-and-Update.md
View File

@@ -29,7 +29,7 @@ Add to your `settings/local.php`:
@define('CONST_Osm2pgsql_Flatnode_File', '/path/to/flatnode.file');
Replace the second part with a suitable path on your system and make sure
the directory exists. There should be at least 40GB of free space.
the directory exists. There should be at least 64GB of free space.
## Downloading additional data
@@ -62,7 +62,7 @@ involve a GB or US postcode. This data can be optionally downloaded:
In its default setup Nominatim is configured to import the full OSM data
set for the entire planet. Such a setup requires a powerful machine with
at least 32GB of RAM and around 800GB of SSD hard disks. Depending on your
at least 64GB of RAM and around 800GB of SSD hard disks. Depending on your
use case there are various ways to reduce the amount of data imported. This
section discusses these methods. They can also be combined.
@@ -124,7 +124,7 @@ The style can be changed with the configuration `CONST_Import_Style`.
To give you an idea of the impact of using the different styles, the table
below gives rough estimates of the final database size after import of a
2018 planet and after using the `--drop` option. It also shows the time
needed for the import on a machine with 32GB RAM, 4 CPUS and SSDs. Note that
needed for the import on a machine with 64GB RAM, 4 CPUS and SSDs. Note that
the given sizes are just an estimate meant for comparison of style requirements.
Your planet import is likely to be larger as the OSM data grows with time.
@@ -135,7 +135,7 @@ street | 42h | 400 GB | 180 GB
address | 59h | 500 GB | 260 GB
full | 80h | 575 GB | 300 GB
You can also customize the styles further. For an description of the
You can also customize the styles further. For a description of the
style format see [the development section](../develop/Import.md).
## Initial import of the data
@@ -147,18 +147,44 @@ Download the data to import and load the data with the following command
from the build directory:
```sh
./utils/setup.php --osm-file <data file> --all [--osm2pgsql-cache 28000] 2>&1 | tee setup.log
./utils/setup.php --osm-file <data file> --all 2>&1 | tee setup.log
```
The `--osm2pgsql-cache` parameter is optional but strongly recommended for
planet imports. It sets the node cache size for the osm2pgsql import part
(see `-C` parameter in osm2pgsql help). As a rule of thumb, this should be
about the same size as the file you are importing but never more than
2/3 of RAM available. If your machine starts swapping reduce the size.
***Note for full planet imports:*** Even on a perfectly configured machine
the import of a full planet takes at least 2 days. Once you see messages
with `Rank .. ETA` appear, the indexing process has started. This part takes
the most time. There are 30 ranks to process. Rank 26 and 30 are the most complex.
They take each about a third of the total import time. If you have not reached
rank 26 after two days of import, it is worth revisiting your system
configuration as it may not be optimal for the import.
Computing word frequency for search terms can improve the performance of
forward geocoding in particular under high load as it helps PostgreSQL's query
planner to make the right decisions. To recompute word counts run:
### Notes on memory usage
In the first step of the import Nominatim uses osm2pgsql to load the OSM data
into the PostgreSQL database. This step is very demanding in terms of RAM usage.
osm2pgsql and PostgreSQL are running in parallel at this point. PostgreSQL
blockes at least the part of RAM that has been configured with the
`shared_buffers` parameter during [PostgreSQL tuning](Installation#PostgreSQL_tuning)
and needs some memory on top of that. osm2pgsql needs at least 2GB of RAM for
its internal data structures, potentially more when it has to process very large
relations. In addition it needs to maintain a cache for node locations. The size
of this cache can be configured with the parameter `--osm2pgsql-cache`.
When importing with a flatnode file, it is best to disable the node cache
completely and leave the memory for the flatnode file. Nominatim will do this
by default, so you do not need to configure anything in this case.
For imports without a flatnode file, set `--osm2pgsql-cache` approximately to
the size of the OSM pbf file (in MB) you are importing. Make sure you leave
enough RAM for PostgreSQL and osm2pgsql as mentioned above. If the system starts
swapping or you are getting out-of-memory errors, reduce the cache size or
even consider using a flatnode file.
## Tuning the database
Accurate word frequency information for search terms helps PostgreSQL's query
planner to make the right decisions. Recomputing them can improve the performance
of forward geocoding in particular under high load. To recompute word counts run:
```sh
./utils/update.php --recompute-word-counts
@@ -176,7 +202,8 @@ you also need to enable these key phrases like this:
./utils/specialphrases.php --wiki-import > specialphrases.sql
psql -d nominatim -f specialphrases.sql
Note that this command downloads the phrases from the wiki link above.
Note that this command downloads the phrases from the wiki link above. You
need internet access for the step.
## Installing Tiger housenumber data for the US

24
docs/admin/Installation.md
View File

@@ -55,15 +55,15 @@ For running continuous updates:
### Hardware
A minimum of 2GB of RAM is required or installation will fail. For a full
planet import 32GB of RAM or more are strongly recommended.
planet import 64GB of RAM or more are strongly recommended. Do not report
out of memory problems if you have less than 64GB RAM.
For a full planet install you will need at least 700GB of hard disk space
For a full planet install you will need at least 800GB of hard disk space
(take into account that the OSM database is growing fast). SSD disks
will help considerably to speed up import and queries.
On a 6-core machine with 32GB RAM and SSDs the import of a full planet takes
a bit more than 2 days. Without SSDs 7-8 days are more realistic.
Even on a well configured machine the import of a full planet takes
at least 2 days. Without SSDs 7-8 days are more realistic.
## Setup of the server
@@ -73,17 +73,21 @@ You might want to tune your PostgreSQL installation so that the later steps
make best use of your hardware. You should tune the following parameters in
your `postgresql.conf` file.
shared_buffers (2GB)
maintenance_work_mem (10GB)
work_mem (50MB)
effective_cache_size (24GB)
shared_buffers = 2GB
maintenance_work_mem = (10GB)
autovacuum_work_mem = 2GB
work_mem = (50MB)
effective_cache_size = (24GB)
synchronous_commit = off
checkpoint_segments = 100 # only for postgresql <= 9.4
max_wal_size = 1GB # postgresql > 9.4
checkpoint_timeout = 10min
checkpoint_completion_target = 0.9
The numbers in brackets behind some parameters seem to work fine for
32GB RAM machine. Adjust to your setup.
64GB RAM machine. Adjust to your setup. A higher number for `max_wal_size`
means that PostgreSQL needs to run checkpoints less often but it does require
the additional space on your disk.
For the initial import, you should also set: