kern/82271: cbq scheduler cause bad latency
Sebastian Koehler
acex5 at syncer.de
Wed Jun 15 14:10:06 GMT 2005
>Number: 82271
>Category: kern
>Synopsis: cbq scheduler cause bad latency
>Confidential: no
>Severity: serious
>Priority: medium
>Responsible: freebsd-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Wed Jun 15 14:10:05 GMT 2005
>Closed-Date:
>Last-Modified:
>Originator: Sebastian Koehler
>Release: 5.4-STABLE
>Organization:
>Environment:
FreeBSD icecube.thrillkill.lan 5.4-STABLE FreeBSD 5.4-STABLE #10: Fri Jun 10 13:52:40 CEST 2005 acex5 at icecube.thrillkill.lan:/usr/obj/usr/src/sys/ICECUBE i386
>Description:
The following altq rule cause a bad latency from all internal clients using my router.
altq on $int_if cbq bandwidth 2Mb queue { std_in, wow_in }
queue std_in bandwidth 128Kb priority 1 cbq(default)
queue wow_in bandwidth 1920Kb priority 4 { wow_std_in, wow_pri_in }
queue wow_std_in bandwidth 1792Kb priority 4 cbq(borrow)
queue wow_pri_in bandwidth 128Kb priority 7 cbq(borrow)
This effect only occurs, when the internal interface is used. If I use priq on internal interface, latency looks good, too.
altq on $ext_if priq bandwidth 2Mb queue { std_out, wow_std_out, wow_pri_out }
queue std_out priq(default)
queue wow_std_out priority 5 priq(red)
queue wow_pri_out priority 6
Without cbq rule on internal interface:
Antwort von 66.249.85.99: Bytes=32 Zeit=17ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=16ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=18ms TTL=247
With cbq rule on internal interface:
Antwort von 66.249.85.99: Bytes=32 Zeit=16ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=18ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=17ms TTL=247
With cbq rule on internal interface and traffic from an internal client (http, online stream etc.)
Antwort von 66.249.85.99: Bytes=32 Zeit=1619ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=4029ms TTL=247
Antwort von 66.249.85.99: Bytes=32 Zeit=1979ms TTL=247
The issue doesn't seems to be architecture related, cause the issue happen on FreeBSD 5.4-STABLE on Sparc64, too.
>How-To-Repeat:
Load the altq rule into pf. Create traffic from an internal client and check latency from same or other host.
>Fix:
n/a
>Release-Note:
>Audit-Trail:
>Unformatted:
More information about the freebsd-bugs
mailing list