Re: Spin bit - semantics - rtt and loss

[<[e-mail]>]

Note that the spin counters don't just use more bits ; they also add 
bidirectional feedback and allow to see loss both downstream and upstream in 
either direction. And their transmission is robust to any level of reordering 
and loss.


On 23/03/2018 08:15, Lubashev, Igor wrote:
> Thanks, Alexandre!
>
> Yes, 7 bits would work 64 times better, but it is 7 times more bits.  That's a difference between adding a byte to short packets vs adding no extra bytes.  I could live with both.
>
> The 10-packet square wave was just an initial suggestion (the exact size is minor details, and I actually expect that to be an endpoint configuration option).  It also seems that having an rtt spin bit alongside the square wave should help in detecting losses that exceed a "wavelength" (20 packets in this case), though someone more analytically talented than me would likely be able to tell exactly what benefits an rtt wave adds to the loss estimation.
>
> - Igor
>
> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]]
> Sent: Friday, March 23, 2018 6:55 AM
> To: Lubashev, Igor<ilubashe=[e-mail]>
> Cc: [e-mail]
> Subject: Re: Spin bit - semantics - rtt and loss
>
> Hi Igor,
>
> Indeed packet loss is important to track.
>
> Your proposal is an instance of the "square sequence" proposed earlier by Kazuho. He suggested a much longer period (hundreds); you're exploring lower values. Each range has issues: too long turns into a large temporal imprecision for sparse flows (like ACKs); too short blurs the signal in case of tail drops (many contiguous losses).
>
> Also, these methods only give upstream losses (between the sender and the midpoint). For this reason we propose a more ambitious 7-bit solution here: Spin Counters
>
>    http://snaggletooth.akam.ai/IETF-101-HotRFC/10-Ferrieux.pdf
>
> Would that help ?
>
> On 23/03/2018 01:12, Lubashev, Igor wrote:
>>  Since we decided to put some work into spin bit, albeit in a separate
>>  document from V1, here are some thoughts.
>>
>>  It is nice that spin bit can be used to estimate rtt. I am sure that
>>  it is useful to many well-meaning people. It is not very useful to me, though.
>>
>>  When I debug network problems, they are much more likely to be about
>>  lost packets. And I like to be able to know packet loss frequency and
>>  pattern very accurately. (I am assuming an encrypted packet number, or
>>  the problem would be trivial.)
>>
>>  I would prefer, if we have a spin bit, for it to spin not every rtt
>>  but every, say, 10 packets that an endpoint sends. Client sends 10
>>  packets with 0 and then
>>  10 packets with 1, repeat. Sever does the same.
>>
>>  So maybe two spin bits -- one for rtt and one for loss? I would assume
>>  that loss bit would assist in rtt estimation in case of loss.
>>
>>  Thoughts?
>>
>>  - Igor
>
>
> _________________________________________________________________________________________________________________________
>
> Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration, Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.
>
> This message and its attachments may contain confidential or privileged information that may be protected by law; they should not be distributed, used or copied without authorisation.
> If you have received this email in error, please notify the sender and delete this message and its attachments.
> As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
> Thank you.
>
>


_________________________________________________________________________________________________________________________

Ce message et ses pieces jointes peuvent contenir des informations confidentielles ou privilegiees et ne doivent donc
pas etre diffuses, exploites ou copies sans autorisation. Si vous avez recu ce message par erreur, veuillez le signaler
a l'expediteur et le detruire ainsi que les pieces jointes. Les messages electroniques etant susceptibles d'alteration,
Orange decline toute responsabilite si ce message a ete altere, deforme ou falsifie. Merci.

This message and its attachments may contain confidential or privileged information that may be protected by law;
they should not be distributed, used or copied without authorisation.
If you have received this email in error, please notify the sender and delete this message and its attachments.
As emails may be altered, Orange is not liable for messages that have been modified, changed or falsified.
Thank you.