[j-nsp] Problem with PB-4GE-SX

[Richard A Steenbergen]

On Thu, Feb 15, 2007 at 08:40:40PM +0100, O. Notka wrote:
> Hi,
> 
> we have two routers, each equipped with one PB-4GE-SX mounted on a
> FPC2-E, but we only get 450 Mbit per port.
> 
> We have already tried to run the port standalone or in ae with 2 ports,
> which then make up to 900 Mbit.
> 
> We use JunOS 7.6R2.6 and 7.0R1.5.
> 
> Has anybody ever heard of such a behaviour or even a solution for that?

Sounds like SFM oversubscription to me. A recent unpublished limitation 
I've become familiar with is the in addition to the stated FPC bandwidth 
constraints there are individual SFM to PIC bandwidth constraints as well.

For example, even in a FPC2 with 3 empty PIC slots, each SFM is only 
capable of providing any individual PIC with only 1 Gbps of fabric 
capacity. So for example if you only have 2 working SFM's, your 4GE pic 
will only have 2 Gbps of RX fabric bandwidth (same thing applies to an 
OC48 too), which fits nicely with your observed limitation once you factor 
in J-Cell overhead. Also, running anything less than 4 active SFM's seems 
to significantly increase the rate of packet reordering on every FPC2 PIC 
in the chassis compared to having all 4 active (at least with FPC non-E's 
which don't support packet scheduling).

Given the rate at which SFM's are dropping like flies due to SRAM errors 
(oh and lets not forget my personal favorite, bad FPCs that can't talk to 
ANY SFM on a certain channel without introducing errors, packet loss, 
crashes, etc), some days it is a real challange to keep M160's working at 
all. The only really safe way is to run all FPC1's or configure M40e mode.

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)