Mirror
/
mergerfs
mirror of https://github.com/trapexit/mergerfs.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1376 Commits
46 Branches
97 Tags
42 MiB
 
 
 
 
Tree: 97e0d1f8ea
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '97e0d1f8ea'
${ noResults }
mergerfs/mkdocs/docs/config/proxy-ioprio.md

2.0 KiB
Raw Blame History

proxy-ioprio

  • type: BOOL
  • default: false
  • example: proxy-ioprio=true

In Linux certain process schedulers have the ability to prioritize based on a per thread IO priority assigned to the software by users using tools such as ionice. Since such details are not provided by the FUSE protocol, users may not use schedulers which support IO priority, and the extra overhead to query the priority and set it within mergerfs it does not by default attempt to mirror/proxy the value. When enabled however, for all read and write requests, mergerfs will query the priority of the requesting process/thread and set the same value on the thread within mergerfs making the read/write.

This may be useful in situation where the system has high IO load or processes are known to have varying priorities. See the man page for specific details.

Keep in mind that if no IO scheduler has been set for a thread then by default the IO priority will match the CPU nice value which for mergerfs is set by scheduling-priority. Nice value proxying may be added in a future release.

Conflicts With Other Options

If using IO passthrough there is no reason to set this value to true. However, since passthrough leads to mergerfs IO calls being bypassed entirely will have no impact on performance or behavior regardless.

When using passthrough the IO priority of the client app would be used directly given the IO is passed through.

Supported Platforms

Only Linux. This is not supported on FreeBSD.

Performance Impact

Despite the additional syscalls requires the impact appears undetectable with larger buffer sizes. At very small buffer sizes (such as 512 bytes) there is a noticable but small impact (~1.5%).