[c-nsp] Cisco 12008 8/40 Gigabit Link
root net
rootnet08 at gmail.com
Thu Sep 13 01:21:28 EDT 2012
On Wed, Sep 12, 2012 at 9:44 AM, Pete Templin <petelists at templin.org> wrote:
> On 9/11/12 9:33 PM, root net wrote:
>
> That's what I am starting to figure out. Been doing some research and
>> it's truly a distributed platform. I've also been reading that we would
>> want Engine 3 as you mention but I think Engine 2 would work too. What
>> do you think?
>>
>
> The Engine 2 cards Suck(tm). Horribly limited code, so you might be able
> to use an ACL, or you might be able to use uRPF, or you might...it's a
> guessing game, and enabling a conflicting feature may cause a brief outage
> while microcode bundles are changed.
Wow! Thanks. Seems like a huge headache would set in.
>
>
> Does that previous employer have two full Internet tables or just one?
>> Is it IPv4 only or both IPv4 and IPv6.
>>
>
> Two tables...bzzzt. Two feeds? Yes. It's all one table. But yes,
> plenty of routers taking multiple feeds (the additional feeds don't take as
> much memory as the first), and they do have v6 as well.
>
>
Yes, of course...I meant feeds not tables. Crazy me...
>
> You are right miss read. Only 124xx has 5xSFC. The 8/40 we are getting
>> has 3xSFC and 1xCSC now. I will add another CSC period for redundancy.
>> It doesn't have any router processors. My choices for Ethernet lines
>> cards seem to be 8-Port FE w/ ECC (engine 1) w/256MB, 1-Port GE w/ECC
>> (engine 1) w/256MB and 3-Port GE (engine 2) w/512MB. Unfortunately the
>> 4-Port GE ISE (engine 3) not supported on the GSR 8/40. That would have
>> been easy enough hands down with a 1GB of route memory on the LC.
>>
>
> Avoid Engine 1 completely. It was hurried to market so they could have
> OC48 or 4xOC12, but it's even worse than E2: no priority queueing is
> possible in the card. See above on the 3xGE; it was worse than the other
> E2 cards I used.
>
> I promise you the 4xGE does work in GSR8/40 aka 12008. I've shoved it in
> and seen it come alive no problem. I think you have to bring it down to a
> lower RAM amount in GRP-B, but it'll work. Previous employer had several.
That would have been interesting to test.
>
>
> If your cards are Engine 3, they'll serve you well up to the
>> 2.5Gbps/slot limits. If your xRP has enough RAM for all of its
>> tables and your linecards have enough RAM for the FIB copy, you'll
>> be fine. I miss these boxes, but they do have a steep learning curve.
>>
>> This 8/40 would have no xRP. So, we would need to decide to go with the
>> GRP-B and upgrade as bandwidth increases.
>>
>
> I think you missed the underlying fact of the GSR architecture: it's all
> about the linecards. A GRP-B could run a fully loaded 12816. Transit
> traffic never touches the RP, so bandwidth is not controlled/limited by the
> RP. Only receive traffic makes its way towards the RP, though some stuff
> is handled on the card. If dCEF dies on a card, the card is out of service
> until dCEF is restarted; there's no "punt to RP" option. (The RP can go in
> any slot, so the RP could only move one slot's worth of traffic if it were
> asked to move traffic; that wouldn't work.)
>
I was actually having an internal thought process of as bandwidth increases
(more money) we could get a PRP for support and should have been more
clear.
But was also thinking that the GRP-B would be borderline anyway and the LC
only being Engine 2 with 512MB max.
Thanks for all the comments.
> pt
>
> http://www.cisco.com/en/US/**products/hw/routers/ps167/**products_tech_**
> note09186a00801365b3.shtml<http://www.cisco.com/en/US/products/hw/routers/ps167/products_tech_note09186a00801365b3.shtml>
> http://www.cisco.com/en/US/**products/hw/routers/ps167/**products_tech_**
> note09186a008015a057.shtml<http://www.cisco.com/en/US/products/hw/routers/ps167/products_tech_note09186a008015a057.shtml>
>
>
>
More information about the cisco-nsp
mailing list