[Bloat] another sets of leaps forward for linux packet queueing

[Bloat] another sets of leaps forward for linux packet queueing

[Dave Taht]

[-- Attachment #1: Type: text/plain, Size: 1150 bytes --]

Eric Dumazet just posted a pure fq scheduler (using the highly optimized
red/black trees in the kernel)

http://marc.info/?l=linux-netdev&m=137740009008261&w=2

which "scales to millions of concurrent flows per qdisc".Jon Corbet wrote
it up in (subscriber only link) in lwn http://lwn.net/Articles/564978/
which also details the new tso sizing patch, which fixes the oft complained
about overlarge tso problem and does much saner things with it....

http://thread.gmane.org/gmane.linux.network/281264

Almost simultaneously Stephen Hemminger wrote a sysctl option making it
possible to easily switch the default qdisc from pfifo_fast to "something
else"

http://marc.info/?l=linux-netdev&m=137764555622344&w=2

And earlier in the week Jesper Bauer got the last of the atm/htb fixes in.

All of these patches are now in the net-next tree. There were a few
touchups and tweaks here and there over the days following, that aren't in
there yet...

Future looks so bright we gotta wear shades, or eyepatches, or something....


-- 
Dave Täht

Fixing bufferbloat with cerowrt:
http://www.teklibre.com/cerowrt/subscribe.html

[-- Attachment #2: Type: text/html, Size: 1757 bytes --]

Re: [Bloat] another sets of leaps forward for linux packet queueing

[Steinar H. Gunderson]

On Sat, Aug 31, 2013 at 01:47:50PM -0700, Dave Taht wrote:
> Eric Dumazet just posted a pure fq scheduler (using the highly optimized
> red/black trees in the kernel)
> 
> http://marc.info/?l=linux-netdev&m=137740009008261&w=2
> 
> which "scales to millions of concurrent flows per qdisc".Jon Corbet wrote
> it up in (subscriber only link) in lwn http://lwn.net/Articles/564978/
> which also details the new tso sizing patch, which fixes the oft complained
> about overlarge tso problem and does much saner things with it....

I think it's hard to overstate the long-term significance of fq and its TCP
pacing; I've done similar things with HTB in the past with excellent results,
so having something non-hackish for this will be awesome. In a few years,
we'll wonder how we ever survived without pacing.

/* Steinar */
-- 
Homepage: http://www.sesse.net/

Re: [Bloat] [Codel] another sets of leaps forward for linux packet queueing

[MUSCARIELLO Luca OLNC/OLN]

[-- Attachment #1: Type: text/plain, Size: 1389 bytes --]




On 08/31/2013 10:47 PM, Dave Taht wrote:
>
>
> Eric Dumazet just posted a pure fq scheduler (using the highly 
> optimized red/black trees in the kernel)
>
> http://marc.info/?l=linux-netdev&m=137740009008261&w=2
>
> which "scales to millions of concurrent flows per qdisc".Jon Corbet 
> wrote it up in (subscriber only link) in lwn 
> http://lwn.net/Articles/564978/
> which also details the new tso sizing patch, which fixes the oft 
> complained about overlarge tso problem and does much saner things with 
> it....
>
> http://thread.gmane.org/gmane.linux.network/281264
>
> Almost simultaneously Stephen Hemminger wrote a sysctl option making 
> it possible to easily switch the default qdisc from pfifo_fast to 
> "something else"
>
> http://marc.info/?l=linux-netdev&m=137764555622344&w=2
>
> And earlier in the week Jesper Bauer got the last of the atm/htb fixes in.
>
> All of these patches are now in the net-next tree. There were a few 
> touchups and tweaks here and there over the days following, that 
> aren't in there yet...
>
> Future looks so bright we gotta wear shades, or eyepatches, or 
> something....
>
>
> -- 
> Dave Täht
>
> Fixing bufferbloat with cerowrt: 
> http://www.teklibre.com/cerowrt/subscribe.html
>
>
> _______________________________________________
> Codel mailing list
> Codel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel


[-- Attachment #2: Type: text/html, Size: 3344 bytes --]

Re: [Bloat] [Codel] another sets of leaps forward for linux packet queueing

[MUSCARIELLO Luca OLNC/OLN]

[-- Attachment #1: Type: text/plain, Size: 1583 bytes --]

Very happy to see this.

However having a look to the code, if I am not wrong,
the drop policy is not longest queue drop.

If codel is not used, at least longest queue drop should be used.

Luca


On 08/31/2013 10:47 PM, Dave Taht wrote:
>
>
> Eric Dumazet just posted a pure fq scheduler (using the highly 
> optimized red/black trees in the kernel)
>
> http://marc.info/?l=linux-netdev&m=137740009008261&w=2
>
> which "scales to millions of concurrent flows per qdisc".Jon Corbet 
> wrote it up in (subscriber only link) in lwn 
> http://lwn.net/Articles/564978/
> which also details the new tso sizing patch, which fixes the oft 
> complained about overlarge tso problem and does much saner things with 
> it....
>
> http://thread.gmane.org/gmane.linux.network/281264
>
> Almost simultaneously Stephen Hemminger wrote a sysctl option making 
> it possible to easily switch the default qdisc from pfifo_fast to 
> "something else"
>
> http://marc.info/?l=linux-netdev&m=137764555622344&w=2
>
> And earlier in the week Jesper Bauer got the last of the atm/htb fixes in.
>
> All of these patches are now in the net-next tree. There were a few 
> touchups and tweaks here and there over the days following, that 
> aren't in there yet...
>
> Future looks so bright we gotta wear shades, or eyepatches, or 
> something....
>
>
> -- 
> Dave Täht
>
> Fixing bufferbloat with cerowrt: 
> http://www.teklibre.com/cerowrt/subscribe.html
>
>
> _______________________________________________
> Codel mailing list
> Codel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel


[-- Attachment #2: Type: text/html, Size: 3609 bytes --]

Re: [Bloat] [Codel] another sets of leaps forward for linux packet queueing

[Eric Dumazet]

On Sat, 2013-08-31 at 13:47 -0700, Dave Taht wrote:
> 
> 
> Eric Dumazet just posted a pure fq scheduler (using the highly
> optimized red/black trees in the kernel) 
> 
> http://marc.info/?l=linux-netdev&m=137740009008261&w=2
> 
> 
> which "scales to millions of concurrent flows per qdisc".Jon Corbet
> wrote it up in (subscriber only link) in lwn
> http://lwn.net/Articles/564978/

Dav, I saw your message on Facebook

Call me on 415-646-5783