4500-X vs 3850 Fiber model in Core

[salemmahara]

Hello everyone

As the followers of these topics might considered, I'm finding answers for non-answered question step by step. I search and search, read and read and find some dark points in documents. So asking questions here will help people to find solutions.

Today, we're going to have a controversial discussion! 4500-X vs 3850 Fiber model, Which one do you prefer to put in Core block ? Why ? . Let's know your answers according to below parameters:

* We are going to compare 4500X-24X-E and 3850-24XS-E

1. Features ( L2,L3,Security, HA, Reliability )

2. Throughput

3. Next generation networks

4. Flexibility and connectivity

5. Hardware and Software support, bugs and.... 

5. Which one do you prefer as your Core switch ? Why?

[Leo Laohoo]

Which one do you prefer as your Core switch ? Why?

Neither.  

I'd go for 6807-X chassis because the supervisor card can be upgraded.  Currently, 6807 can support Sup2T and Sup6T.  I suspect, in the next 18 to 24 months, the monster Sup10T will be released.  

Sup2T can support up to 220 Gbps per slot while Sup6T is 440 Gbps per slot.  Sup10T will support 880 Gbps.  If I want to protect my investment, 6807 is my choice.  

[salemmahara]

Hi Leo

Thanks for replying

Leo, 6807s need 10 units in data center. If we decide to have a redundant core, we need one more. So we'll need 20units! we don't have this much space in our racks right now and resizing the data center might not be admitted by board of directors because we're going to move to new building in next 5 years.

So, do you still choose 6807 or a cheaper core switch? I guess choosing 6807 for a network which is going to be regenerated in next 5 years is not reasonable. Is it? I mean, If we buy 6807 and pay a lot of money on it, we probably have to use it in new building so our core switch wouldn't be replaced for 10 years(maybe).

Imagine, You're going to setup a redundant core with 6807 . Could you give me your idea about chassis and SUPs and modules?  (Regarding to our copper backbone )

On the other hand, our core switch must do below task :

1. Run EIGRP with 3 other building ( Each one has 6-7 floors and a stack of 3850 as core to inter vlan routing ).

2. Handle its building ( 4 floors with 200 user at all )

3. Handle server farm connectivity to internet and LAN ( Our server farm switch is a 2960G and our core is 2960G again right now - we have a flat L2 network with a boradcast domain of 4 building) . I'm here to make everything alright and my target is a L3 network with many little Broadcast domain. BTW, out 2960Gs aren't under any pressure with this situation . So buying a couple of 6807 with some extra SUPs might be a waste of money. Isn't it?

Leo, I'm going to appreciate you for your kindness and patience . 

[Leo Laohoo]

If we buy 6807 and pay a lot of money on it, we probably have to use it in new building so our core switch wouldn't be replaced for 10 years(maybe).

The 6807 solution is one unique chassis-based system.  Similar to the 4500 and the Nexus 7K/9K.

The single chassis of a 6807 can support different models of supervisor cards.  Currently, Sup2T and Sup6T are supported.  In the next 18 to 24 months, the next "monster" will come out, Sup10T.  This supervisor card can support up to 880 Gbps per slot bandwidth.  

The price you pay for the chassis will return in dividends.  Sup6T is aimed at 100 Gbps links (up to 4 per slot) while the Sup10T will be aimed at high density 40 Gbps and eight 100 Gbps links.  

That alone translates to me that if ain't protecting my investment, then I don't know what is. 

Another thing, 6807, 6880 and 6840 support a feature called Instant Access. 

To answer your question, yes, if I want a monster core switch that can (and will) handle 100 Gbps, then 6807-X is my solution.  

[salemmahara]

Thank you again Leo But did you think of cost? A couple of 3850-24xs-e VS a 6807 with redundant Sup 6T or 2T and of course port card! 2 unit rack VS 10 Unit rack? Redundant PSU,FAN and .... cost too.

Are you sure about your choice for the network which I described? Your solution triple the cost ( At least - without redundant chassis )

[rasmus.elmholt]

Is it still the case for the IA that all traffic between to local IA ports is backhauled? This would be for me a problem if all intra-port traffic needs to go over the uplinks.

[Leo Laohoo]

A server farm being served with 2960G spells disaster.  Catalyst switches are known to have very shallow port buffer.  This translate to slow transfer of speed to and from the servers.  This is why Cisco Nexus family of switches are doing a booming trade due to it's niche position of doing nothing but server-facing role.  

[att-csc-aia]

These 2 models have similar costs by adding things up.

I'll go for 4500X if I don't need more than 2 units stacking together.

Just blindly believe that Cisco positions 4500X higher than 3850 for a reason. lol 

[rasmus.elmholt]

I would agree to this, afaik the 3850 does not support ISSU yet and is a bit flaky i my experience. Wouldn't but the 3850 as my DC core for those reasons.

[Joseph W. Doherty]

The 4500-X is "industrial strength" vs. a 3850.

Rather than either, you might also look at the 6840-X and/or Catalyst 9300 (3K successor) and 9500 series.