[j-nsp] ISSU offlined mpc - why?

[Saku Ytti]

On Wed, 1 Sept 2021 at 20:35, Chuck Anderson via juniper-nsp
<juniper-nsp at puck.nether.net> wrote:

> Eventually during the ISSU process, the line card software needs to be
> upgraded.  During that part, each line card goes offline one at a
> time.  If you have multiple line cards, design your network such that
> redundant network paths are connected across different cards to
> prevent a total outage when each line card is upgraded one-by-one.

Disclaimer: I do not use ISSU nor do I plan to use it, it is complex
and I've been hurt before with non-obvious failure modes after the box
is left in an unknown state, this is vendor agnostic fear I have and I
am currently not seeking help to overcome it.

https://www.juniper.net/documentation/us/en/software/junos/high-availability/topics/topic-map/unified-issu-enhanced-mode.html
Enhanced mode is an in-service software upgrade (ISSU) option
available on MPC8E, MPC9E, and MPC11E line cards that eliminates
packet loss during the unified ISSU process

The changes in hardware that affect MPC behaviour are mostly the
microcode that the lookup engines run, in trio these are collections
of identical cores called PPE (packet processing engine). In theory we
could put one PPE at a time out of service, and restart it with new
ucode, until we're all done, having N-1/N of nominal PPS capacity
during upgrade (I think EA has 96 PPE, so 98.95% during upgrade),
And in fact this is what 'hyper mode' does, Trio has ucode compatible
history of over a decade, without running hyper mode the PPEs are
running the old +decade old collection of code, with hyper mode the
PPEs are running new rewrite of the ucode. Newer hardware is hyper
mode only, older hardware it is operator choice which generation of
ucode to run.

-- 
  ++ytti