[Codel] Worst case conditions - large RTT

[Codel] Worst case conditions - large RTT

[Simon Barber]

Codel assesses the minimum delay over a fixed time period (the interval) 
- what happens when the interval is much smaller that the RTT. For 
example a tcp stream is experiencing a 2 second RTT (perhaps due to 
bufferbloat at some point), yet the interval is 100ms. The TCP 
implementation is a poor one that will ACK a window at a time. Here 
codel will drop excessively, and cause low throughput?

Simon

Re: [Codel] Worst case conditions - large RTT

[Dave Taht]

On Wed, May 9, 2012 at 1:36 PM, Simon Barber <simon@superduper.net> wrote:
> Codel assesses the minimum delay over a fixed time period (the interval) -
> what happens when the interval is much smaller that the RTT. For example a
> tcp stream is experiencing a 2 second RTT (perhaps due to bufferbloat at
> some point), yet the interval is 100ms. The TCP implementation is a poor one
> that will ACK a window at a time. Here codel will drop excessively, and
> cause low throughput?
>

You are doing a great job of asking good questions.

I would hope we could get some more people lined up to answer them,
armed with sims and real hardware. Volunteers for each problem would
be helpful.

Me, I'd like very much to get some data back on the variable bandwidth
wireless scenario! Except that cerowrt's too broken at the moment to
use and I'm trying to fix that.

The latest netem does ECN, btw. And ns3 has a decent minstrel model.
(hint, hint)

> Simon
> _______________________________________________
> Codel mailing list
> Codel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel



-- 
Dave Täht
SKYPE: davetaht
US Tel: 1-239-829-5608
http://www.bufferbloat.net

Re: [Codel] Worst case conditions - large RTT

[Jim Gettys]

On 05/09/2012 04:41 PM, Dave Taht wrote:
> On Wed, May 9, 2012 at 1:36 PM, Simon Barber <simon@superduper.net> wrote:
>> Codel assesses the minimum delay over a fixed time period (the interval) -
>> what happens when the interval is much smaller that the RTT. For example a
>> tcp stream is experiencing a 2 second RTT (perhaps due to bufferbloat at
>> some point), yet the interval is 100ms. The TCP implementation is a poor one
>> that will ACK a window at a time. Here codel will drop excessively, and
>> cause low throughput?
>>
>
One test I'd love to see is to compute from first principles what the
latency and jitter we should now be observing in a single ideal TCP
flow, at several RTT's.  Then we can compare with that latency and
jitter we *actually* observe.

I'd expect a slightly higher value right now, as BQL's buffering is
being added to the queue that CoDel is managing, but we can at least see
if the implementation we have is behaving exactly as we would predict.
                    - Jim

Re: [Codel] Worst case conditions - large RTT

[Kathleen Nichols]

In the paper, I show the 500 RTT case. I also ran some at 750ms and 1 sec.
The utilization is not as good as the smaller RTTs but it was still good.
Maybe you can try to create the situation you are thinking of?

On 5/9/12 1:36 PM, Simon Barber wrote:
> Codel assesses the minimum delay over a fixed time period (the interval)
> - what happens when the interval is much smaller that the RTT. For
> example a tcp stream is experiencing a 2 second RTT (perhaps due to
> bufferbloat at some point), yet the interval is 100ms. The TCP
> implementation is a poor one that will ACK a window at a time. Here
> codel will drop excessively, and cause low throughput?
> 
> Simon
> _______________________________________________
> Codel mailing list
> Codel@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/codel