[j-nsp] Separate internet transit network versus converged

[Saku Ytti]

On 29 March 2016 at 15:05, chip <chip.gwyn at gmail.com> wrote:
> Juniper has a "This Week" book freely available that discussess, in detail,
> the path of a packet through an MX.  It's quite an informative read.
>
> http://www.juniper.net/us/en/training/jnbooks/day-one/networking-technologies-series/packet-walkthrough-mx-series/

I love that 'few milliseconds', only 3 orders of magnitude off.

> This Week: An Expert Packet Walkthrough on the MX Series 3D provides the
> curious engineer with a global view of the short life (a few milliseconds)
> of packets inside the Juniper Networks MX Series of 3D routers.

Technically not JNPR book, but reverse engineered by JNPR customer,
which is bit of ironic that we have to do that. There is great
internal Trio document (now dated) by Steven Wong. But it is really
really good for external customers too. Who knows what gems they have
I don't have any idea, which would have helped me avoid downtime or
bad design choices.
I would put in all future RFQ/RFP requirement for internal-level
architecture documentation, since getting these otherwise is super
hard. I don't understand why vendors don't publish these. Mentioned
document has helped me countless time, to choose design which least
exposes platforms compromises, to troubleshoot issue without bothering
JTAC or helped me give JTAC precise troubleshooting information
potentially cutting weeks of solution time and downtime of repeated
problems.

Lot of the good instrumentation is not exposed to end-users at all,
like packet-via-dmem or capturing exception packets (you're getting
CRC errors from IXP, which neighbour is to blame?, you're getting IP
checksum errors from MPLS core, who is mangling them?). Traditionally
very hard to answer questions become trivial with supplied
instrumentation.

-- 
  ++ytti