Re: Explicit measurability in the QUIC wire image (was Re: Packet number encryption)

[Kazuho Oku <[e-mail]>]

2018-02-08 11:08 GMT+09:00 Kazuho Oku <[e-mail]>:
> 2018-02-08 9:37 GMT+09:00 Christian Huitema <[e-mail]>:
>>
>>
>> On 2/7/2018 1:10 AM, Kazuho Oku wrote:
>>> The pros of using a sequence that randomly distributes 0 and ones
>>> (i.e. PI or xorshift+) is that you can extract good loss / reordering
>>> information if the error rate is low.
>>>
>>> OTOH, it is fragile to high error rate or certain patterns of error.
>>> Let me explain two examples.
>>>
>>> Consider the case where the packet loss rate is 50%. I doubt if you
>>> could use pattern matching against those sequences to figure out the
>>> loss rate. This is because the power among the sequence is mostly high
>>> frequencies. If you use a sequence that distributes power among lower
>>> frequencies (like square wave or sin wave that I discussed), you can
>>> calculate the loss rate in such cases as well.
>>
>> Actually, it should be possible to build a maximum likelihood decoder
>> for an xorshift sending pattern. You are mostly concerned with erasure
>> and permutations, and you roughly know the range of probability, so you
>> can see likelihood of 1, 2, 3, etc. erasures, or permutations in some
>> range. Obviously, the range that you can manage depends on the length of
>> the code. If you can keep the decoder synchronized, you will be able to
>> measure the erasure rate.
>
> I agree with you that it is possible to detect loss / reorder if the
> decoder can be kept synchronized.
>
> What I am questioning is the possibility of keeping the decoder
> synchronized under heavy loss / reorder.
>
>> If it looses synchronization, you know that
>> the error rate is truly horrendous, and that's information too.
>
> I disagree.
>
> It might be true that loosing synchronization due to heavy loss can be
> a signal. OTOH, loosing synchronization due to massive reorder is not
> an issue if the reorder resolves in a certain timeframe (IIRC 1/8
> RTT).
>
> My understanding is that the middleboxes' reliance to the sequence
> number has led to the ossification of TCP. The ossification has
> prevented us from doing some nifty things, such as sending some TCP
> packets through a different node while answering others directly. In
> QUIC, I am interested in doing such things, and I would expect that
> others will be.
>
> Considering that, I would prefer not to exposing a signal that could
> give false information to observers when there is a "good" reorder.

Please let me paraphrase the last sentence to the following:

I would prefer using a signal that is tolerant to "good" reorder (i.e.
happening frequently but resolving in a fraction of RTT so that they
don't trigger a retransmit).



Anyways, I think that we are making progress in sense that we are
discussing what is the "best" way to achieve loss / reorder
observability using 1 bit, rather than using multiple bits.

Use of less bits lead to less chance of leaking information as well as
finding space to store the information.

>
>> And
>> xorshift codes are self synchronizing, so you will regain
>> synchronization quickly.
>>
>> -- Christian Huitema
>>
>
>
>
> --
> Kazuho Oku



-- 
Kazuho Oku