Re: [Tsv-art] Tsvart early review of draft-ietf-lsvr-l3dl-03

[Joerg Ott <[e-mail]>]

No worries, I am happy about any break I get...

On 23.05.20 20:32, Randy Bush wrote:
> sorry to take a while to get back to this.  your review is really
> appreciated.
> 
> i think we converged on most things.  the following seemed open.
> 
>>>> 3. When the protocol applies fragmentation, should there be a note on
>>>> preventing bursts?
>>>
>>> likely part of this is our fault, as we did not mean 'fragmentation' in
>>> the classic "oops!  we found a hop with a small mtu."
>>
>> I didn't take it to mean classic fragmentation but rather ALF-style
>> operation.  Still, this could generate bursts depending on how much
>> information there is to 'fragment'.
> 
> yes, it is app level framing.  perhaps we should call it that explicitly
> or even segmentataion or some term less well known.
> 
> do you perhaps have a specific suggestion?

Not really.  This all appears artificial if you need two or three
packets for app layer fragmentation.  Maybe one could write something
substantially improved along the lines of:
To prevent packet bursts, a sender SHOULD pace the transmission of
application layer fragmented data units as follows: A sender MAY
transmit up to K packets containing fragments in a burst and SHOULD
pace bursts ... (but how?)

>>>> Section 7 on the checksum needs more detail.
>>>
>>> could you be more specific re in what area?
>>
>> Let me go back:
>>
>> The code is fine (and certainly a good idea).  I was kind of expecting
>> the code to be described in text, but given the code is there, the
>> concise description is fine.  Except that I got confused by
>> "repeat until zero".  What should be zero?  Looking at the code, surely
>> not the result.
> 
> :)
> 
>      Sum up 32-bit unsigned ints in a 64-bit long, then take the
>      high-order section, shift it right filling on the left with zeros,
>      rotate, add it in, repeat until the high order 32 bits are all zero.

Sold :-)

>>>> Also wondering if repeated retries (due to failure, not lost packets)
>>>> could yield fast repeated transmissions.
>>>
>>> could you be specific about what you think could be improved in
>>
>> I was trying to decipher my handwriting on the tablet, and it appears to
>> say "wait before retry to prevent fast loops", scribbled on the bottom
>> of p.22.  I had to go back tot he text and I believe this refers back to
>> p.18, first para:
>>
>> "The Key is specific to the operational environment.  A failure to
>>     authenticate is a failure to start the L3DL session, an ERROR PDU
>>     MUST BE sent (Error Code 3), and HELLOs MUST be restarted."
>>
>> If you have a systematic auth failure, this would cause HELLOs to be
>> restarted.  Without any jittering (now you introduced some above),
>> you could, in principle, get a loop:
>>
>> HELLO -> OPEN -> (Auth) ERROR -> HELLO -> OPEN -> (Auth) ERROR -> ...
>>
>> So misconfig could be bad unless there is some delay built in.
> 
> so i added
> 
>      Although delay and jitter in responding with an OPEN were specified
>      above, beware of load created by long strings of authentication
>      failures and retries.
> 
> but i am unsure of what action to recommend.

Count to N, raise an alert, pause.  Or something like this?

>>> pre-published text and xml are at https://git.rg.net/randy/draft-l3dl
> 
> still holds true

Thanks -- best,
Jörg