[Bloat] bufferbloat on 10Gig fiber at the gathering?

[Bloat] bufferbloat on 10Gig fiber at the gathering?

[Dave Taht]

[-- Attachment #1: Type: text/plain, Size: 1435 bytes --]

see a rrul test on this link at - (sorry you will have to unmangle the
url, no dns for this yet)

http:// 151 dot 216 dot 125 dot 10/germany/rrul/long_period_swings.svg

I am curious if this is an artifact of the data, the rrul test itself,
or actually observing long period latency swings due to bufferbloat at
these enormous speeds.

Similarly, I did a long period test to my machine in california:

http:// 151 dot 216 dot 125 dot 10 /~d/california/postfixes_long.svg

The measured latency on this link, unloaded, was 195ms, earlier this
morning before 5000 participants arrived

http:// 151 dot 216 dot 125 dot 10 /~d/rrul/to_california2.svg

so there appears to be ~15ms of intrinsic latency added, and then
anywhere between 10-200ms more, at the bottleneck fiber link to the
internet from "the gathering", too.

plenty of other data on the same ~d url, I figure the above format
will pass your spam stoppers.

There is no RED nor WRR configured on this cisco box, and although it
appears both are supported on the 10gig line card, I have no idea how
to configure RED at these speeds (3 10gig fiber lines, 2 of which are
basically saturated so far). Suggestions?

Very interesting data thus far. I am getting tons of captures. It is a
really impressive scene, see my g+ account for some pictures.

-- 
Dave Täht

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

[-- Attachment #2: bloaton10gigfiber.png --]
[-- Type: image/png, Size: 254508 bytes --]

Re: [Bloat] bufferbloat on 10Gig fiber at the gathering?

[Mikael Abrahamsson]

On Wed, 27 Mar 2013, Dave Taht wrote:

> There is no RED nor WRR configured on this cisco box, and although it 
> appears both are supported on the 10gig line card, I have no idea how to 
> configure RED at these speeds (3 10gig fiber lines, 2 of which are 
> basically saturated so far). Suggestions?

What platform is used? ASR9000? It has ~50 ms of FIFO buffering per port 
if nothing is configured if I remember correctly.

policy-map EGRESS
class class-default
random-detect precedence 0 10 ms 30 ms

int teng0/0/0/0
service-policy EGRESS output

This should create a policy-map called EGRESS which does WRED for prec 0 
traffic with 0% drop until 10 ms and 100% drop at 30ms of buffer depth, 
and then this is attached in the egress-direction on tengi0/0/0/0

> Very interesting data thus far. I am getting tons of captures. It is a 
> really impressive scene, see my g+ account for some pictures.

I'm kind of disappointed that they don't have enough bw there, lately 
these kinds of events have had more bandwidth available than has been 
used because ISPs and vendors have used them as tech demonstrators, and 
also that people have enough bw available at home so it's been less 
downloading/uploading done there.

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se

Re: [Bloat] bufferbloat on 10Gig fiber at the gathering?

[Mikael Abrahamsson]

On Thu, 28 Mar 2013, Mikael Abrahamsson wrote:

> What platform is used? ASR9000? It has ~50 ms of FIFO buffering per port if 
> nothing is configured if I remember correctly.

According to <http://technet.gathering.org/page/2/> it's 6500 so my guess 
is that the buffering you're seeing isn't within the LAN but instead 
within the ISP network, because the 6500 linecards don't really have that 
much buffer space.

> policy-map EGRESS
> class class-default
> random-detect precedence 0 10 ms 30 ms
>
> int teng0/0/0/0
> service-policy EGRESS output

This won't work on Cisco 6500.

-- 
Mikael Abrahamsson    email: swmike@swm.pp.se