Re: [tcpm] alternate IW proposal

[Joe Touch <[e-mail]>]

On 11/18/2010 10:06 AM, Mark Allman wrote:
>
>>>     - The cross connection dynamic is presuming you're making the same
>>>       mistakes over-and-over.  That isn't clear.  If you make a zillion
>>>       10 packet connections to some host then OK.  I am willing to call
>>>       that pathological.
>>
>> Let me be more clear. The IW isn't a property of the destination,
>> it's a property of the path.
>
> No it isn't.  The IW is a construct of the sender.  The path DOES NOT
> HAVE A WINDOW, LET ALONE AN **INITIAL** WINDOW.  Period.

Granted. The path doesn't have an IW. The path supports a given IW.

> Congestion is a property of a particular link.

Congestion is a property of a path. It can occur at a link, or across a 
set of links (e.g., if parallelized, or alternated). It can also occur 
on a sequence of links, e.g., if you send 1 Gbps through a path that 
includes links with remaining capacity of 100 Mbps and 10 Mbps, they 
BOTH experience congestion-based loss.

...
> So, we use some default IW.  That default IW is a probabilistic
> engineering *guess* at its root.  We use something that we believe will
> work well in the vast majority of the cases and when it doesn't the
> results are not somehow catastrophic.
>
>      (It *could* have a history as you have noted (and conducted research
>      on).  But, the practical efficacy of that has at least not been
>      shown (as far as I know; perhaps I am out of the loop).  E.g., how
>      often do two hosts communicate?

You keep assuming that whether an IW works is a property of a single 
source, or of a pair of endpoints.

I am concerned about cases where the IW interacts with a link that 
affects communication with MANY hosts, and so ends up being a property 
of a set of receivers, or all receivers.

>> How much does this help? How much does it reduce latency? Most of
>> the info I've seen summarized suggests the benefit is in the 5-20%
>> range. That's not even noticeable from a user POV.
>
> 20% seems like a pretty nice chunk of change to me.  That's on the order
> of what we got by going to 4KB.  20% of a second is perceptible to
> humans.

Really? You can tell the difference between 1 second and 0.8 seconds? or 
1.2 seconds? Or 400 vs 500 ms?

 > Add this over a number of objects and it looks like a fine bit
> of increase to me.

It doesn't generally add. Separate connections are independent; the only 
way it adds is when you retrieve something in a subsequent connection 
based on the first, at which point you could have used persistent 
connections and avoided the TWHS, which saves an additional 2 RTTs *and* 
allow you to use your current window (which could be larger than the IW).

So basically the only time it adds up is when you should be doing 
something different for a more significant benefit anyway.

Joe

Joe