[c-nsp] Small PE L3 Switches ?

Mike Butash der.mikus at gmail.com
Fri Sep 15 02:41:00 EDT 2006


Depends on your business model, but really it tends to come down to a 
6500 if you're talking port density and massive throughput.  7300's, 
both the 7301 and 7304 work good too for most circumstances, but are 
only good if you're talking about needing a few interfaces.  We couldn't 
quantify the cost in moving to a BFR like a GSR or CRS for even some 
very large networks, but I don't have to route terabytes across a wan 
quite yet.  The 6500ME and 3750ME's have some supposed promise too, but 
I've not worked with these yet.

My two cents:

7304 with nse (mpls hardware accelerated) you get 2 gige's per nse, so 
ethernet doesn't go far if you're talking a lot of bandwidth are going 
to be delivered.  If you're talking pos, atm, or other tdm, then it's 
not so bad.  There are weird quirks with these using items such as an 8 
t1 pa card in mlppp with all of them will kill the box because it 
process switches it all, nor can it bundle t1's in an 8 port IMA for an 
equally weird and dysfunctional implementation.  Use of 2x gige's 
doesn't go far when you're talking about redundancy, there goes your 
uplinks, now you need other ports, so you're talking SPA adapters at 
something like 20k for 4 gige's.  This gets expensive fast.

7301 is just a 2600 on steroids with a pa card slot in it and 3x gige's. 
Nothing notable, but not a bad device.  Good for multiple t1's and don't 
share the same behavior as it's cousin, the 04.

6500 we'd found makes for a nice mpls p or pe platform, though there are 
limitations in some cases where they are less than ideal.  For ethernet, 
loading a 6500 with 6748's or 6704 10g blades is king of capacity, and 
can act functionally as a PE great.  You have to understand the 
distributed architecture and limitations within the hardware combo's as 
they pertain to features and functionality within the platform.

For qos, they don't do any shaping, wred, or some other class-map 
features you find in routers (short of an OSM or SIP with it's own 
processing) when packets never leave a line-card dfc or punted to the 
pfc, but instead rely on asic queuing to get the job done though 
wrr-queueing and mls qos functions.  They can remark packets just fine 
for cos/tos/dscp/exp, policing, mutation, and rely on bit boundary 
mapping between the types for asic tail-drop prioritizing.  Best bet for 
maintaining sanity is to have a cohesive enterprise-wide mapping for 
cos/dscp/exp and the 6500's are quite functional.

They do all the major routing protocols, and offer fairly advance 
capability such as mpls te, vpls, queue-based tunnel selection, and 
other goodies that should keep you happy.  Most of it is very stable if 
you don't ask too much of each box for duties and roles it performs. 
SRA code offers some nice promise of features once customers do enough 
Q/A for Cisco and it gets stable.

Modules are pure value-add too, such as fwsm's and everything else they 
have for them now.  Intro price is a bit steep for sup720's, but they 
are excellent platforms to operate if designed properly, and can scale 
quite impressively.  I just wish they'd come up with some sane cable 
management for the 6748's, these get damn ugly in time once populated. 
Think MTP/MPO connectors now cisco, they work well enough for the DWDM 
platforms.  :)

-mb



David J. Hughes wrote:
> On 14/09/2006, at 7:49 PM, sthaug at nethelp.no wrote:
> 
>> Part of the Cisco strategy of confusing the customer as much as
>> possible. That's why you have
> 
> 
> And just to round out the list with a router you have
> 
> - 7301 and 7304 which other than the Cisco logo have nothing in common.
> 
> You are right.  Damn confusing.  And it doesn't help me find a new PE  
> product :-)
> 
> 
> 
> David
> ...
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
> 


More information about the cisco-nsp mailing list