[email protected]
[Top] [All Lists]

RE: composite link - candidate for respin, maybe

Subject: RE: composite link - candidate for respin, maybe
From: John E Drake
Date: Mon, 29 Mar 2010 18:01:23 -0700
Tony,

Clearly there is extra overhead (roughly 2*(N - 1)* 40 bytes?) with N 
homogeneous bundled links versus one heterogeneous bundled link.  The concern I 
have is that having heterogeneous bundled links may complicate the data plane 
packet forwarding.

A clear statement as to what we are trying to accomplish wrt IGP scaling by 
using bundled links would be good.

Thanks,

John

> -----Original Message-----
> From: Tony Li [mailto:[email protected]]
> Sent: Monday, March 29, 2010 2:12 PM
> To: John E Drake
> Cc: [email protected]; [email protected]
> Subject: Re: composite link - candidate for respin, maybe
> 
> 
> Hi John,
> 
> > The difference between advertising one bundled link with multiple sets
> > of parameters or multiple bundled links each with one set of
> > parameters doesn't seem so significant.
> 
> 
> Well, that depends entirely on our scalability goals, which is why I would
> like to see our requirements quantified.
> 
> As an example, if we use IS-IS with TE right now, a link consists of:
>       2 octets of TLV overhead
>       7 octets of system Id and pseudonode number
>       3 octets of default metric
>       1 octet of length of sub-TLVs
>       6 octets of Administrative group (color)
>       6 octets of IPv4 interface address
>       6 octets of IPv4 neighbor address
>       6 octets of Maximum link bandwidth
>       6 octets of Reservable link bandwidth
>       34 octets of Unreserved bandwidth (for 8 traffic classes)
>       5 octets of TE Default metric
> 
> If we want to track latency, then I'd propose that we need 3 bytes to
> track
> worst case latency on the link  (2 bytes overhead, 1 byte for ms
> granularity
> latency).
> 
> If I've done my math right, that works out to 85 bytes.
> 
> If we then vectorize this, we would need, per component:
>     4 octets of maximum link bandwidth
>     4 octets of reservable link bandwidth
>     32 octets of unreserved bandwidth
>     1 octet of latency
> 
> That's 41 bytes.
> 
> If we're willing to dispense with pre-reserved bandwidth per traffic
> class,
> then we can further drop 32 octets.
> 
> So we're down to 9 bytes per link, or almost a 10x improvement in
> scalability.
> 
> IMHO, that's significant.
> 
> If we now have ~16 components per composite link, then we can advertise
> the
> composite link in about 200 bytes.  That works out to well over 1000
> composite links (and therefore 16,000 components) per system.  And that's
> before we break a sweat and do funky things to get more LSP space (e.g.,
> expanded MTUs, zero-cost pseduonodes for more fragments).
> 
> So, what are the real requirements?
> 
> Tony
> 

_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

<Prev in Thread] Current Thread [Next in Thread>