[j-nsp] MX204 vs. MX240??
adamv0025 at netconsultings.com
adamv0025 at netconsultings.com
Thu Nov 14 08:19:07 EST 2019
> Saku Ytti
> Sent: Tuesday, November 12, 2019 11:28 AM
>
> On Tue, 12 Nov 2019 at 12:34, Mark Tinka <mark.tinka at seacom.mu> wrote:
>
> > On our MX480's, we've known for a very long time that IPv4 BFD is
> > supported in the PFE. However, IPv6 BFD runs on the RE.
>
> PFE is an ambiguous term, it variably means NPU or LC CPU inside JNPR.
>
> There are several places where you can run your keepalieve
>
> a) RPD
> b) RE PPMd
> c) LC CPU PPMd
> d) NPU (dispatch block in the LU/XL)
>
> And it depends on config where you run it. Not every protocol can register
to
> all of these. But yes, BFD v4 can run as RE PPMd, LC CPU PPMd and NPU.
>
Hey Saku,
Would need some more insight from you to process the above
I must admit I got confused by the above since I was only familiar with
this:
https://kb.juniper.net/InfoCenter/index?page=content&id=KB31595&pmv=print&ac
tp=LIST&searchid=&type=currentpaging
Right so RPD or whoever offloads the periodic packet management to PPMD that
part is fine.
Then PPMD can run on RE obviously or can be distributed onto LC -now this is
where my confusion comes in.
That's because I was under the impression that it's the LC CPU hosting the
PPMD in distributed mode,
But reading the KB article linked above again and other sources on PPMD also
specifically mentions PFE as the entity running PPMD
And in in my book PFE is solely the Trio complex (MQ,LU,QX -or whatever
version and quantity combinations of thee 3 blocks), but not the LC CPU.
-hence my confusion,
And yes I didn't do a test of running say 3k BFD sessions in PPMD
distributed mode to see if LC CPU utilization or PFE utilization rises.
But still I don't see how the LU microcode has the ability to actually
generate packets, let alone to host a complete daemon.
Alright the PPEs inside LU are pretty much generic cores (probably some
streamlined instruction set) but could potentially run any type of app, but
the microcode I find it hard to believe it has enough code to run a daemon
or generate packets -what about hyper-mode (streamlining of that microcode)
-would that have any effect on the ability of the microcode to run PPMD?
In my head I find it much more plausible for the LC CPU just tx/rx packets
to/from trio complex and the trio just doing the simple
forwarding/filtering/queuing/counting..
Also is there a cmd I can use to switch between LC CPU and LU?
Thanks,
adam
More information about the juniper-nsp
mailing list