[Iccrg] Fwd: agenda planning for ICCRG in Anaheim

[Matt Mathis]

On Sun, Feb 21, 2010 at 6:08 PM, Lachlan Andrew
<lachlan.andrew at gmail.com> wrote:
> Greetings Michael
>
> On 22 February 2010 09:07, Michael Welzl <michawe at ifi.uio.no> wrote:
>> On Feb 21, 2010, at 10:48 PM, Matt Mathis wrote:
>>>
>>> Therefor: all stacks should include a secondary congestion control
>>> mechanism that detects the delay caused by large queues in the network
>>> and regulates their congestion window accordingly.
>>
>>> It may be an appropriate future work item of the ICCRG and IETF to
>>> attempt to standardize such a mechanism.   In the short term it would
>>> be sufficient to merely agree on a statement of principle or intent.
>>
>> - attempting to standardize: isn't that exactly what LEDBAT is doing?
>
> I thought LEDBAT was specifically for less-than-best-effort.  Matt's
> proposal is that *all* stacks should consider delay.  That is, we also
> want a replacement which is competitive with standard TCP "on
> average", although perhaps less aggressive when buffers are large and
> more aggressive either in the presence of non-congestion loss or when
> recovering from transient congestion.

Even simpler: use the same basic algorithm as LEDBAT, except use a set
point that is double the baseline delay plus a small constant.   This
basically stops the queue from getting too much larger than necessary
to maintain full link rate following any background random loss.

My home ISP has 5 full seconds of queuing on the outbound side.  Given
that standard TCP controls against a full drop-tail queue and modern
stacks autotune out to a couple of megabyte windows, simply
downloading a large file makes the link absolutely unusable for
anything else.   Although one could argue that the real problem is
with the ISP, there is simply no reason for TCP to create a standing
queue that is 50 times larger than the baseline delay.

I didn't realize the potential scale of the impending trainwreck,
until one of my adult children commented about how his nice new
computer completely trampled everything else on his home LAN.    He
asked if I had any idea why.....

I think we have only seen the tip of the iceberg.

Thanks,
--MM--