[LARTC] Recent changes in qdisc/cls APIs

[LARTC] Recent changes in qdisc/cls APIs

[Thomas Graf]

Folks,

I've been changing some of the classifier APIs especially bits
related to the rate estimator and statistics between 2.6.8
and 2.6.10. This probably broke some external classifiers
such as esfq. If you happen to use one of them just send
me the correspondig patch and I will fix it up for you.

Cheers
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Andy Furniss]

Thomas Graf wrote:
> Folks,
> 
> I've been changing some of the classifier APIs especially bits
> related to the rate estimator and statistics between 2.6.8
> and 2.6.10. This probably broke some external classifiers
> such as esfq. If you happen to use one of them just send
> me the correspondig patch and I will fix it up for you.
> 
> Cheers

Thanks for the warning - I can't give a patch until I remember where I 
got it/ what I've done.
I had to do some by hand anyway as it looks like the new iproute uses 
it's own copy of some kernel files.

I use overlimits as a counter in esfq - has this changed ?

Andy.


_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Thomas Graf]

> I use overlimits as a counter in esfq - has this changed ?

I assume you're using your own tc_stats copy and not abusing
the policer stats. You may want to use the new statistic
interface which provides overlimits via gnet_stats_queue.
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Andy Furniss]

Thomas Graf wrote:
>>I use overlimits as a counter in esfq - has this changed ?
> 
> 
> I assume you're using your own tc_stats copy and not abusing
> the policer stats. You may want to use the new statistic
> interface which provides overlimits via gnet_stats_queue.
> 

I used overlimits as it was always 0 for qdisc - it doesn't seem to 
affect things, but then I don't know whether anything else uses 
individual qdisc stats. It's called from HTB whose overlimits counters 
seem unaffected.


Andy.

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Andy Furniss]

Andy Furniss wrote:
> Thomas Graf wrote:
> 
>>> I use overlimits as a counter in esfq - has this changed ?
>>
>>
>>
>> I assume you're using your own tc_stats copy and not abusing
>> the policer stats. You may want to use the new statistic
>> interface which provides overlimits via gnet_stats_queue.
>>
> 
> I used overlimits as it was always 0 for qdisc - it doesn't seem to 
> affect things, but then I don't know whether anything else uses 
> individual qdisc stats. It's called from HTB whose overlimits counters 
> seem unaffected.

Are these the only changes needed -

http://www.kernel.org/diff/diffview.cgi?file=%2Fpub%2Flinux%2Fkernel%2Fv2.6%2Ftesting%2Fpatch-2.6.10-rc1.bz2;z&61

Andy.

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Thomas Graf]

* Andy Furniss <418A4B3E.9070801@dsl.pipex.com> 2004-11-04 15:31
> Are these the only changes needed -
> 
> http://www.kernel.org/diff/diffview.cgi?file=%2Fpub%2Flinux%2Fkernel%2Fv2.6%2Ftesting%2Fpatch-2.6.10-rc1.bz2;z&61

Yes and no. It should work, however I suggest to adapt to the new API
completely:

If you happen to have xstats:
http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_qdisc_dump_stats.diff

If classful:
http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_gen_stats.diff
http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_dump_stats.diff
http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_rate_est.diff

And preferely introduce requeue statistics:
http://people.suug.ch/~tgr/patches/accepted/gen_stats/requeue_stats.diff

If you happen to dump the stats yourself, remove them:
http://people.suug.ch/~tgr/patches/accepted/gen_stats/remove_bogus_qdisc_stats_copy-26.diff

As I said, I can do those changes for you if you want, it's a matter of
a few minutes for me.
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Re: [LARTC] Recent changes in qdisc/cls APIs

[Andy Furniss]

Thomas Graf wrote:
> * Andy Furniss <418A4B3E.9070801@dsl.pipex.com> 2004-11-04 15:31
> 
>>Are these the only changes needed -
>>
>>http://www.kernel.org/diff/diffview.cgi?file=%2Fpub%2Flinux%2Fkernel%2Fv2.6%2Ftesting%2Fpatch-2.6.10-rc1.bz2;z&61
> 
> 
> Yes and no. It should work, however I suggest to adapt to the new API
> completely:
> 
> If you happen to have xstats:
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_qdisc_dump_stats.diff
> 
> If classful:
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_gen_stats.diff
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_dump_stats.diff
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/cbq_class_rate_est.diff
> 
> And preferely introduce requeue statistics:
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/requeue_stats.diff
> 
> If you happen to dump the stats yourself, remove them:
> http://people.suug.ch/~tgr/patches/accepted/gen_stats/remove_bogus_qdisc_stats_copy-26.diff
> 
> As I said, I can do those changes for you if you want, it's a matter of
> a few minutes for me.
> 

Thanks for all those - I'll need time to digest them and won't be 
upgrading just yet.

My esfq is hacked in other ways aswell so a patch against what people 
really use will be better - if someone else wants to post. The iproute 
bit needs changing aswell for those that use the latest versions.

As for using overlimits - It was just for a test - I saw an unused 
counter which showed up in stats and used it for my own purpose.

I incremented it if the queue wasn't empty so I could tell if bulk 
traffic got into my interactive class. I'll probably get rid of it.

Andy.

_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/