Add documentation about system requirements and suggestions#4937
Conversation
optionsome
added
the
!Documentation
|
I think CI is affected by #4929 |
Codecov ReportPatch coverage has no change and project coverage change:
Additional details and impacted files@@ Coverage Diff @@
## dev-2.x #4937 +/- ##
=============================================
+ Coverage 63.02% 63.15% +0.12%
- Complexity 13370 13406 +36
=============================================
Files 1669 1669
Lines 66505 66519 +14
Branches 7253 7250 -3
=============================================
+ Hits 41917 42009 +92
+ Misses 22192 22111 -81
- Partials 2396 2399 +3
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. ☔ View full report at Codecov. |
Co-authored-by: Leonard Ehrenfried <mail@leonard.io>
| OTP's performance scales with the number of available CPU cores. OTP processes each request in a separate thread and depending on the configuration, a single request can use multiple threads. Realtime updates also run in a separate thread. Therefore, to have a good performance, it makes sense to have multiple cores available. How OTP uses parallel processing also depends on the available cores (<= 2 cores vs >2 cores) in some cases. Therefore, load testing should be done against a machine that doesn't differ too much from production machines. | ||
|
|
||
| Entur and the Digitransit project have found that the 3rd generation AMD processors have a slightly better performance for OTP2 than the Intel 3rd generation CPUs (and especially better than the 2nd generation CPUs). |
There was a problem hiding this comment.
Maybe emphasize that one OTP http request runs for most parts in one thread. There are small operations
part of an request that can be configured to run in parallel, but in general we do not utilize more than
one core per request. However, this is something we continuously evaluate and may change.
There was a problem hiding this comment.
Legacy GraphQL API allows batch queries (and queries containing multiple queries) to be processed totally in parallel if the queries are plan/nearest queries, If configured to do so. But you are correct that almost always we only use singly thread (at least for the most part of the query) so I'll change the wording.
t2gran
left a comment
t2gran
left a comment
There was a problem hiding this comment.
Only one comment about threading.
Co-authored-by: Leonard Ehrenfried <mail@leonard.io>
3 participants
Summary
Adds documentation page for the system requirements and suggestions.
Issue
No issue
Documentation
Updated documentation
Changelog
From title.