Re: call for testers: altq in current

From: Eygene Ryabinkin <rea-fbsd_at_codelabs.ru>
Date: Mon, 9 Apr 2007 13:40:10 +0400
Nate, good day.

Mon, Apr 09, 2007 at 05:56:14AM +0000, nate_at_root.org wrote:
> A few weeks ago, I committed a change to ALTQ that I was only able to
> compile-test.  What I need is someone with a laptop or other
> cpufreq-capable system that is also using ALTQ to verify that with
> powerd running, the queuing timing is now reliable.

I see no difference between the -CURRENT from today and from 30th
March (I see that your commit was made at 26th of March, but I am
not sure that mu current was updated after it for the -CURRENT
compiled at 30th of March).

The bad news are that the ALTQ behaves wrong: when the CPU frequency
is changed the bandwidth changes too. Either I am doing something
wrong, or your commit should be polished a bit.

My environment is: Asus A2D running AMD Mobile XP, iwi (Intel
2915ABG) and 7-CURRENT. The pf rules were:
-----
altq on iwi0 bandwidth 3Kb cbq queue { dflt }
queue dflt bandwidth 100% cbq(default)
pass out quick log on iwi0 proto tcp from $my_ip to any flags S/AUSPF \
keep state queue dflt
-----

The interface is running at 22 Mbit/sec most of the time. No polling
was enabled. The bandwidth was measured by the ifstat, powerd was
disabled and I had changed the frequency via sysctl. Four frequencies
were used: 400, 800, 1600 and 2200. The kernel config included the
following ALTQ options:
-----
options		ALTQ
options		ALTQ_CBQ
options		ALTQ_RED
options		ALTQ_RIO
options		ALTQ_HFSC
options		ALTQ_CDNR
options		ALTQ_PRIQ
-----
Basically, I made two tests: WAN and LAN downloading over FTP and
HTTP.  WAN test was done for the old and new -CURRENTs and LAN test
was done just for the new -CURRENT. All tests were done in the
following manner: ifstat was spawned with the delay of 3 seconds,
files were downloaded by fetch and I manually changed the CPU
frequency via sysctl.

First two logs, ifstat.bw3Kb.old.wan.log and ifstat.bw3Kb.new.wan.log
do show the WAN results. The 100 Kbps corresponds to 400 MHz, 200
Kbps -- to 800 MHz, 410 Kbps -- to 1600 MHz and 560 Kbps -- to 2200
MHz CPU speed. I thought that I was bounded by the WAN link here.

ifstat.bw3Kb.new.local-net.log shows the behaviour for the LAN link
(machines are sitting on the same switch, so only L2 devices are
on the network path): just the same figures and bandwidth is still
changing with the CPU.

ifstat.fullbw.new.local-net.log shows the behaviour for the 100%
bandwidth, but with the ALTQ rules enabled (the '3K' were just
changed to '100%' in the pf.conf). The speed is pretty stable flying
around 20 Mbit/sec.

And ifstat.new.local-net.altq-disabled.log shows the behaviour of
interface bandwidth for pf.conf without any ALTQ rules. The same
as above: 20 Mbit/sec, the interface bound.
-- 
Eygene

Received on Mon Apr 09 2007 - 07:40:22 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:08 UTC