3850 DL - Stacked vs cascaded

[Rajagopal Sivaramakrishnan]

Hi

Looking at the specs of 3850, it looks a good fit for DL's for small sites.. QOS, multicasting etc characteristics looks decent.. we are in a process of deciding 3850 stacked vs 3850 cascaded DL design.. 

Option 1 -  With 3850 stacked we have 2 x 3850's virtually looking as a single DL layer.. We will not have HSRP or STP issues.. we can configure etherchannels from AL switches and have the links active/active..

Option 2 - 3850 cascaded back to back will go back with our legacy design with HSRP/STP   - this looks more neat with layer 2 between the 3850s.. we are more used to this design..

Has anyone seen any issues with 3850 stacked in Distribution layer ?  I presume there arent any , but which design model would you choose ? why ?

Note - we are running a full fledged QOS (Diffserve) network.. also using multicast in our network. VRF lite is a mandatory requirement.

Raj

[Jon Marshall]

Raj

Haven't used them myself so i will only comment on the general query about the design.

I too am very used to the option 2) version where you have HSRP/STP but i think with stacking/VSS on 4500/6500s etc. this is becoming the new model to use. The advantages are -

• no manual load balancing of HSRP active with STP root so you can use both uplinks for every vlan
  
• full uplink capcity from access switches. Even with manual load balancing as in the above you are still not using full uplink capacity because a vlan can only forward on one link
  
• Ease of configuration ie. no need to replicate things between the switches as you would if they were not stacked
  
• With the legacy L2 design you are reliant on STP as an active particpant in the topology. With a stacked solution STP still runs but it is relegated to the background ie. it is not needed to block or forward on switch interconnects so improved convergence times. 
  

When i was designing networks in the past you had to choose between -

1) L2 access to L3 distro (interconnected via a L2 trunk)  where vlans could exist on multiple switches. This is where you start load balancing with HSRP active/STP root etc.

2) L2 access to L3 distro (interconnected via a L3 routed link). With this design vlans should not be allowed to exist on multiple access switches and so both uplinks could be used but having the vlan only allowed on one access switch could be limiting

3) L3 routed access to L3 distro. With this design it was even more limiting ie. it wasn't that vlans should only be on one access switch, they had to be.

Of the 3 design options above 1) was always the most flexible but at a cost of more configuration, reliance on STP etc. Now with the new model you can have the flexibility of 1) with most of the cost removed.

I am struggling to think of a reason not to do it.

Having said all that as per your note about QOS, multicast and VRF lite, i cannot see a reason why a stacked solution would not meet the requirements whereas standalone interconnected switches would but like i say i have not used these switches so perhaps others could comment.

Jon

[Rajagopal Sivaramakrishnan]

Jon

Thanks, as usual

Im rarely using my old id "sachinraja" .. if you remember me

If it was 4500/6500, then we could have thought about VSS, but with 3850 (like 3750), we introduce the stack layer.. when I draw this on a visio diagram, the design looks pretty ordinary with 1 DL switch and various AL's (though logically there are 2 DLs)... Im not against it, but we are so used to square or triangle topology..

Infact stacking gives us more options - like u said - with not running active/standby trunks, and using the uplinks better.. QOS, multicast, vrf would work good too...

Im still confused

Raj

[Jon Marshall · Accepted solution]

Raj

I do indeed remember you. I hope things are good with you.

Im not against it, but we are so used to square or triangle topology..

I know exactly what you mean, it takes an adjustment to get used to it. But i think it is worth the effort because i think the advantages far outweigh the disadvantages and it does give you more flexibility in terms of how you do your design.

If you understand VSS and are comfortable with that just think of it as a similiar thing in terms of design.

I do know what you mean about drawing it though as it would look like just 2 switches connected together

Jon

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Haven't gotten to try the 3850s (they're still relatively new), but have used 3750 stacks for all kinds of purposes, since Cisco came out with the original series.

Jon has well noted many of the pros/cons of the stack vs. distinct units.  One issue with a stack (or VSS), the stack's OS becomes a single point of failure.  So, in theory, you can obtain a better MTBF staying with distinct "parallel" devices.  However, the probability differences are so small, many take the advantages of the stack over the minimal loss in MTBF.  (As possibly the most likely reason for an "IOS failure", is a configuration screw up, you need to be even more careful with change management.)

One disadvantage of the stack, IOS upgrades, as the whole stack goes off-line for a bit (but newer IOSs are supposed to address that with rolling-upgrades).

Another advantage of the stack, often overlooked, it flattens your topology, so whatever you're doing, often both L2 or L3 is "smaller".  "Smaller" often provides faster convergence and perhaps improved stability (less nodes to slosh about).

Oh, and stacks often provide much more inter-member bandwidth.

[Rajagopal Sivaramakrishnan]

Hi Joe

Thanks for your valuable inputs

"One issue with a stack (or VSS), the stack's OS becomes a single point of failure" - are you more talking on the IOS bugs etc ?

One disadvantage of the stack, IOS upgrades, as the whole stack goes off-line for a bit - I think this is a valid point ill check how 3850 is done.

Thanks & Regards

Raj

[Collin Clark]

My 2 cents....I have deployed quite a few 3850's and for the most part I like them. As far as a distribution switch....maybe. The 3850's are not as stringent on design as the 3750's, with a faster CPU, 64-bit OS, etc. The 3850 is still an access layer switch and I use them as a small collapsed core, but a network of decent size, I would look strongly at the 4500-X. It uses VSS so you still get redundnacy whiile eliminating STP & HSRP.

I agree the drawings are harder to draw, but I usually use shaded icons for redundancy

[Rajagopal Sivaramakrishnan]

Collin

Yes.. these are small/medium sites.. some of these sites are running 3750 as DL, but as standalone switches (not stacked)...

3850 is so new.. first thing - Im not sure if there will be too many bugs on routing layer, qos, multicasting etc which are core functions of DL..  apart from the hardware specs, my confusion is stacked DLs vs standalone DLs... have u implemented 3850's as stacked DLs ?

Raj

[Collin Clark]

I only used stacked DL's. I don't see a downside to using them stacked. I can understand your concern about the IOS being buggy however, the 3850's run IOS-XE, and I believe most of the L3 code was ported from the ASR code. I have found them to be very reliable and have not had any issues yet. The high CPU and memory utilization that's prevalent in the 3750-X's is gone.

[Rajagopal Sivaramakrishnan]

That gives more confidence Collin

Thanks

[Rajagopal Sivaramakrishnan]

Thanks John

and good to know u remembered me.. its been a while since i contributed to netpro.. its great to see you are still an active member here... I salute your dedication and passion..

A square design in standalone DL, would become a triangle design with Stacked DL.. I hence think it would always be better to have the uplinks from ALs directly connect to the DLs , which will enable us to configure etherchannels etc with stacks..

Standalone DL

DL1 ----- DL2

|              |

|              |

AL1-------AL2

the same version in stacked design would look like:

______ DL_____

|                       |

AL1   ---------  AL2

Spanning tree again plays a big role with such designs.. If I think of stacked DLs I would rather of a triangle standalone design which would loook like this with stacks:

       DL -------- AL1

         |______AL2

Ill have to convince the architects   hopefully I do

Raj

[Collin Clark]

Logically boxes and triangles are going away. We now build linear lines (or try really hard too). Physically we still make triangles-no boxes!

[Rajagopal Sivaramakrishnan]

Yeah.. this what I was referring to.. I think we should forget square topology with stacked design...

Raj

[Collin Clark]

Correct, do not box with stacked design!

[Jon Marshall]

Raj

its great to see you are still an active member here... I salute your dedication and passion..

Much appreciated although i have to be honest and say i have been away from the forums for quite a while and only just come back

Jon