[c-nsp] Products for dealing with packet re-ordering?

heasley heas at shrubbery.net
Wed Jan 10 18:02:16 EST 2018


Tue, Jan 09, 2018 at 06:05:15PM +0200, Saku Ytti:
> Hey Alex,
> 
> The daemon is not exaggerated, but it's also not inherent problem to
> TCP. TCP inherently guarantees correct order, so reordering isn't an
> issue to TCP on itself. Over time we've noticed that networks are
> lossy but retain order, so we've tinkered with TCP algorithms so that
> it has become very adapt in dealing with packet loss but at the time
> wholly dependent on network keeping order, as misorder is immediately
> interpreted by common TCP algorithms to imply a packet loss, so resend
> is requested.

is that entirely true? misorder is not (should not be) immediately
interpretted as loss.  it does imply loss and might encite an
implementation to ACK or duplicate ACK a preceding segment, but both
sides should wait a reasonable time based on observed RTO for
retransmission.

so, if the implementation doesn't stink and the jitter isnt too bad ...

> If you control both ends already, you could just change TCP algorithm
> to one which tolerates reordering and all would be well. There is also
> somewhat accidental light at the end of the tunnel, due to various
> inefficiencies, including being too aggressive and bursting during
> window growth lot of the content shops are looking to change to new
> TCP algorithm called BBR, it just so happens it'll be more tolerant to
> reordering as it's somewhat different in detecting packet loss.
> 
> 
> On 9 January 2018 at 17:51, Alex K. <nsp.lists at gmail.com> wrote:
> > Hello everyone,
> >
> > An interesting issue came across my desk, one I'd like to hear community
> > opinion about.
> >
> > A customer of mine, due to carrier restrictions, is about to aggregate a
> > few links into one (using etherchannel). Due to required bandwidth, it
> > seems he would need to load balance across the LAG on a packet by packet
> > basis. Now, we all know the dangers of packet reordering. On another hand,
> > there are products from VeloCloud, Talari, SilverPeak and probably many
> > others, that on paper are capable of dealing with the issues and delivering
> > packets in orderly manner.
> >
> > The question is, is anybody using those products and can admit first-hand
> > experience? Or the daemon of packet reordering is grossly exaggerated?
> >
> > Thank you for sharing your experience/opinions on the matter.
> > _______________________________________________
> > cisco-nsp mailing list  cisco-nsp at puck.nether.net
> > https://puck.nether.net/mailman/listinfo/cisco-nsp
> > archive at http://puck.nether.net/pipermail/cisco-nsp/
> 
> 
> 
> -- 
>   ++ytti
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/


More information about the cisco-nsp mailing list