[Bloat] AQM usage in current Internet routers

[Bloat] AQM usage in current Internet routers

[Haiqing Jiang]

[-- Attachment #1: Type: text/plain, Size: 571 bytes --]

Dear Alll,

I read the blog from Getty, saying that RED or other variants of AQM
actually are not ran in current Internet routers.
Could you send me the statistics evidence about that?
If not, do you know how to detect / evaluate the usage of RED?

For example, how to detect whether RED or other AQM versions are used in
routers? Maybe it's a good work to revisit
AQM deployment in the wild of the Internet? What's your opinion??? Thanks,

-- 
-----------------------------------
Haiqing Jiang, PH.D student

Computer Science Department, North Carolina State University

[-- Attachment #2: Type: text/html, Size: 756 bytes --]

Re: [Bloat] AQM usage in current Internet routers

[Oliver Hohlfeld]

Hi,

> I read the blog from Getty, saying that RED or other variants of AQM
> actually are not ran in current Internet routers.

That's a myth. RED is /not widely/ used, as it's configuration is
unclear and subject to a debate that has never converged. But there
are large operators out there that do deploy RED in their backbone,
so saying that it's never used is simple incorrect.

However, I also would be interested in statistics on the deployment
of AQM.

> If not, do you know how to detect / evaluate the usage of RED?

There is related work out there. See for example the paper by
Dischinger et al.:
http://www.mpi-sws.org/~mdischin/papers/07_imc_bb.pdf
See for example Figure 16 and Section 4.2.2.

> Maybe it's a good work to revisit AQM deployment in the wild of the
> Internet? What's your opinion???

But what would be the message of this work? I guess providing some
number would be nice, but would they have any impact? What if you
figure out that RED is not widely deployed (which is known); would it
make a difference to know the exact number for a subset of ASes? To me,
the pros and cons of RED are known and so are the reasons why or why
not to deploy it.
However, it could be interesting to track the deployment of Coddle and
see if operators adapt due to its ease of use. But that would be a
longitudinal study.

Best,
Oliver