Modify

Opened 2 years ago

Closed 2 years ago

Last modified 2 years ago

#21326 closed defect (fixed)

Small fq_codel quantum considered harmfull

Reported by: fuller Owned by: developers
Priority: normal Milestone: Designated Driver (Trunk)
Component: kernel Version: Trunk
Keywords: fq_codel quantum cpu Cc:

Description

i ran into cpu limits with sirq on archer c7v2 while testing ingress shaping.

it seems to me that quantum < mtu is very costly (at cpu) with higher (50mbps) bandwidths

~40% sirq load with quantum 1522+
90% at quantum mtu/2

iirc this was somewhere on ietf-aqm list, but cant find it

maybe it would be wise to undo https://dev.openwrt.org/changeset/36302/

Attachments (0)

Change History (3)

comment:1 Changed 2 years ago by fuller

ie. sqm-scripts and others may set quantum explicitly for low bw links and it is not a good generic optimisation, especially as it sivers down to the wifi codels

comment:2 Changed 2 years ago by nbd

  • Resolution set to fixed
  • Status changed from new to closed

thanks for bringing this to my attention. fixed in r47811

comment:3 Changed 2 years ago by dtaht

well, it IS of benefit for stuff running at lower rates like 10mbit native, or less (say, dsl).

However so little seems to be running at anything but shaped rates down there, at the moment at least, that I support reverting this to the default quantum.

cake tries to autotune the quantum.

Add Comment

Modify Ticket

Action
as closed .
The resolution will be deleted. Next status will be 'reopened'.
Author


E-mail address and user name can be saved in the Preferences.

 
Note: See TracTickets for help on using tickets.