Single router for P/PE functions

Erik Schmersal nanog at schmersal.us
Fri Sep 4 14:07:34 UTC 2009


Hi dave,

Our setup was a dual ring with two devices common to both rings. It used a
full mesh of LSP's but the majority of traffic was L3VPN. There were some
VPLS connections as well, maybe a total of 30 VLAN's. LSP's were set up with
static path's the short way around the ring and a standby active secondary
path the long way around. Convergence time for a failure on either ring was
barely noticable. I am no longer with that organization so I can't get
access to the gear anymore :(

>From my experience, you are probably just asking the EX4200 to do more than
it was made to do. That is a lot of CCC circuits to reallocate on the fly,
especially for a smaller device. You may me able to reduce convergence time
by making your LSP's static with a standby secondary so the path is
preconfigured when a failover occurs, the only problem with this is the
scalability gets poor quickly as you start to add devices.

Erik


On Fri, Sep 4, 2009 at 8:39 AM, daveb <spike at zitomedia.net> wrote:

>
> Hi, I saw your response on NANOG and have a couple questions for you if you
> don't mind.  I'm doing a similar design with MPLS (OSPF/RSVP) on EX4200s in
> a 10GE ring, mainly for 'ccc' circuits and IP connectivity.  The EX4200
> serves both the P and PE functions and some of my rings may be as large as
> 30 devices.
>
> In my informal lab test with just 4 EXs in a ring, the convergence time
> (optomized with FRR, path protection and 50ms BDF) for 1 ccc circuit was
> 300ms and with 200 ccc circuits it was several seconds, and 800 kills the
> box.  I can't imaging what it would be like with 20 or 30 device in the
> ring.
>
> I was just wondering if you've doen similar testing with the MX as far as
> scaling.  I'm assuming the EX4200 just isn't up to the task but I'm also
> concerned that ring topologies are problematic for re-routing LSPs.  I can
> test to find the optimum/maximum number of allowable ccc circuits with 4
> devices in the ring but I have no way or testing with 20 or 30 so I'm really
> trying to determine how much worse convergence is with more devices vs
> number of LSPs.
>
> Thanks,
> Dave Bernardi
>
>
>
> At 12:00 AM 9/4/2009, Erik Schmersal wrote:
>
>> > Not only can they, it's done quite frequently. I just completed a
>> >> deployment of seven Juniper MX series routers in a dual ring
>> configuration,
>> >> all acting as combination P/PE devices for a state government WAN
>> backbone.
>> >> Works like a charm.
>> >>
>> >> Erik
>> >>
>> >>
>> >> On Thu, Sep 3, 2009 at 10:20 AM, Serge Vautour <sergevautour at yahoo.ca
>> >wrote:
>> >>
>> >>> Hello,
>> >>>
>> >>> I'm pretty confident that a router can be used to perform P & PE
>> >>> functions simultaneously. What about from a best practice perspective?
>> Is
>> >>> this something that should be completely avoided? Why? We're
>> considering
>> >>> doing this as a temporary workaround but we all know temporary usually
>> lasts
>> >>> a long time. I'd like to know what kind of mess awaits if we let this
>> one
>> >>> go.
>> >>>
>> >>> Thanks,
>> >>> Serge
>> >>>
>> >>>
>> >>>
>> >>>
>>  __________________________________________________________________
>> >>> Yahoo! Canada Toolbar: Search from anywhere on the web, and bookmark
>> your
>> >>> favourite sites. Download it now
>> >>> http://ca.toolbar.yahoo.com.
>> >>>
>> >>>
>> >>
>> >
>>
>
>



More information about the NANOG mailing list