[c-nsp] ASR9001 Vs ASR1006

James Bensley jwbensley at gmail.com
Thu May 19 03:36:01 EDT 2016


On 19 May 2016 at 00:56, Adam Vitkovsky <Adam.Vitkovsky at gamma.co.uk> wrote:
>> Gert Doering
....
>> Hardware redundancy is totaly overvalued - our 6500s could all run dual-
>> SUPs, and except for the layer2 things, *none* have dual-SUPs - because for
>> a L3 scenario, just putting a second box next to it, running a different IOS
>> train (!) will give you much better resiliency than "dual SUPs, same operating
>> system, crashing together", or "dual SUPs, something dies in a weird way
>> blocking backplane traffic"...
>>
>> Now, for a customer edge box, terminating 10.000s of customers, I can see
>> that you want dual-RSP and all that - plus reasonable reboot-free software
>> upgrades, so maybe not ASR9k either...
>>
> That's actually a good point.
> From what I know 10K customers (VPN not subs) on a single box would be way too many eggs in one basket (doesn't matter the box is fully redundant).
> So yeah it can so happen that you'd have big 10 slot chassis which is half empty cause business won't allow you to terminate more customers on a single box.
> And also it's a question of RSP resources as well -like you might run out of BGP sessions before you'll run out of ports or LCs
> So yeah I guess more of smaller boxes not necesarly 2RU non redundant but just not 10 or 22 slots might be the right way forward.


Yes, this is something I am battling with at $dayjob of late. In the
access/aggregation layer the largest (highest port capacity/line card
capacity) boxes we have are 6500s and at the PE layer it is
7600s/ASR9000s/MX480s/MX960s.

At the PE Layer none of the ASR9006s, MX480s or MX960s are completely
full, only some of the 7606-S are full (not the biggest variety in
that chassis range either). At the access aggregation layer some of
the smaller 6500 chassis are full, no 6513-Es for example are full
though.

I'm having the same arguments with management of late, we can't pile
(read "I don't think we should") any more customers on to these
chassis. Not from a technical perspective but from a business
perspective. The impact from doing an IOS upgrade (for example) on a
fully loaded larger chassis like a 6513 or MX960 (so it will affect
many customers) should be mitigated by customers taking a resilient
service (such as connectivity to two different PoPs or PEs, or by
being connected to PoP sites with dual PEs etc) however even for
customers that pay less for a non-resilient service, watch them jump
and scream when they get more than N number of planned maintenance
works within a given time period.

More smaller boxes works well in my opinion. It comes with other
benefits like not having to have loads of high end (costly) boxes in
the lab.


Cheers,
James.


More information about the cisco-nsp mailing list