[j-nsp] MX104 RE performance

[Blake Willis]

On Sun Apr 13 2014, Gavin Henry wrote:

> Any news yet?

MX104 completed our dual-stack BGP convergence test approximately twice as fast 
as MX80.  Both boxes were running 13.2R4.

Note that the RPD scheduling "fix" for PR836197 
(http://www.nanog.org/meetings/nanog57/presentations/Tuesday/tue.lightning2.steenbergen.juniper-slowfib.pdf) 
slowed convergence down a lot; MX104 under 13.2R4 runs our test in about the 
same time (maybe a little faster) as MX80 under 11.4R7.

If you're looking seriously at the platform, you should ask your Juniper SE to 
PoC one with your use-case.

> Am looking at a new MX5 with upgrade licenses vs this with 2 x 10G

The new 2x10GE + full L3 bundles are commercially quite interesting.  The only 
less expensive MX is a base MX5 (if you don't need 10GE).

	However, one thing I found to be braindead on the MX104 was the lack of 
dual flash chips or dual-root partitioning.  You simply can't snapshot on an RE 
without a USB key stuck in the front.  ACX has a similar architecture & had 
dual-root from FCS (and softupdates turned on too).  Hopefully journalled 
softupdates will make this less painful one day, but it's still a glaring 
omission (Juniper's continued lack of a useful AUX port on anything other than 
the now-EoL JSR is also annoying, but remains merely an annoyance).

JTAC confirmed that this single-root architecture was an active engineering 
decision, so if you're looking at MX104 and want to be able to "request system 
snapshot" on this "hardened" platform without a USB key, then make sure your 
salesperson & SE know about it...

---
 Blake Willis
 Network Engineering Consultant
 blake at 2112 dot net

  "Education enabling individuals to overcome their reluctance or inability to
take full advantage of technological advances and product innovation can be a
means of increasing economic opportunity."

   --Alan Greenspan