Replies: 1 comment 3 replies
-
This doesn't make sense. There is zero chance HDDs are giving you 5-7GB/s speeds. The max theoretical speed of a SATA 3 connection is just 6Gbit/s. Your dd command is not correct and allowing for caching.
This should not be surprising at all. The whole point of nullrw is to make actual read and writes not happen. The data is thrown away on a write and returns success. There is no interaction with underlying branches. The point of the feature is to test theoretical max perf. What the overhead of FUSE / mergerfs is.
Your testing methodology is wrong. Please show what you're doing with dd. You should just use the examples I provide in the docs. The cache.files=off values look totally normal and are in line with the likely actual performance of the drives. |
Beta Was this translation helpful? Give feedback.
-
Hi,
I'm trying to troubleshoot why i'm getting significantly slower read and write times against the mergerfs mount path than via the branch path.
Setup:
Currently, the 'best' performance i seem to be able to achieve is by setting
cache.files=off
. Which gives me about 235 MB/s against the mergerfs mount.When i run the dd commands against the branches directly, i'm getting ~5-7GB/s read and ~230MB/s write.
Interestingly, when using
nullrw
the write speed against the mergerfs mount shows as being faster than to the branch mounts.Current mergerfs configuration
Prior to this testing, i had all of the cache options set to their defaults and was getting roughly the same read/write speeds (~235MB/s Read&Write) that i get with cache.files=off.
I've done some testing with different configuration options:
nullrw=true
cache.files=off
cache.files=partial
cache.files=libfuse
Any suggestions for configuration changes to improve my read speeds would be awesome, if they help with write speeds as well thats a bonus!
Beta Was this translation helpful? Give feedback.
All reactions