Choosing between 4500 & 6500

[branfarm1]

Hi there,

I currently run a pair of 4507R's with SupV's as the core of my datacenter network. All of my servers are dual homed either to the dual 4507's or to HP Blade switches inside HP blade enclosures. I'm trying to decided if I want to upgrade to the -E series 4500 or if I should make the jump to 6500. I know everybody loves the 6500 and would recommend that in a heartbeat, but I need some real reasons why the 6500 would be better. Since i have no experience with 6500's I thought I'd turn here for advice. The important things about my environment are:

--100 servers currently, with plans to expand

--I process a lot of real-time multicast and unicast traffic

--All of the users are connected via the WAN (extranets, direct p2p connections, internet)

-- Low Latency and high availability are the most important design considerations

--Easy of administration

So far, I've been looking at a 6506-E with either a Sup 32 or Sup 2 with SFM and MSFC2. It seems to me that everyone is buying the Sup 32, but when I read the docs it looks like the Sup 2 has better traffic handling capabilities.

Can anyone offer any advice?

Thanks in advance!

[Edison Ortiz]

Supervisor IIs are no longer available, your options are Sup32, Sup32-PISA, Sup-720 or Sup-720-10G.

For Core datacenter I recommend getting at least a Sup-720 with DFC enabled modules.

The throughput achieved with this hardware configuration dramatically surpasses to what you are getting with a 4507.

For more information on the 6500 architecture, please refer to this white paper:

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper0900aecd80673385.html

HTH,

__

Edison.

[Joseph W. Doherty]

Some of the major differences between the 4500 series and the 6500 series, the latter can provide much more performance and many more features. (Note even more features are available with the 6500 since it also supports various service modules that install within the slots.)

Sup2 and SFM might now be EOL (end of life). The advantage of the sup2 (15 Mpps bus - 30 Mpps fabric) with SFM (256 Gbps), is it offers dual 8 Gbps fabric connections per slot, while the sup32 provides only the 32 Gbps bus (and 15 Mpps) per chassis. (For comparison, your 4500 supV provides 6 Gbps fabric per slot, 96 Gbps and 72 Mpps).

With the 6500, since the fabric is "optional", 6500 line cards are designed to work with only the 32 Gbps bus, the 32 Gbps bus and one 8 Gbps fabric connection, dual 8 Gbps fabric, or 1 or 2 connections to the 20 Gbps fabric (provided by the sup720 [720 Gbps] variants).

None of the 6500s sups, alone, offer forwarding performance that matches your 4500 supV. However, some 6500 fabric cards support local processing on the card (provided by a DFC). Cards with the installed DFCs provide an aggregate performance much higher than the 4500 supV.

For chassis supervisor redundancy, for both the 4500 and 6500 you, of course, need dual sups, but the 4500 series requires special versions of the chassis, the "R" variants.

For chassis redundancy, the 6500 provides the Virtual Switching Supervisor Engine 720 which allows dual 6500s to function as one.

For 100 servers, or more, and assuming gig connections, I believe the 4500's limitation of 6 Gbps per slot is insufficent. Likely Cisco believes so to, since there's the newer -E 4500 series variants (chassis, line cards, supervisor) which provides 24 Gbps per slot.

PS:

A bit of a wildcard for 100 plus servers, might be the multi-unit stack of 3750-Es. These offer about wire speed bandwidth and pps per 3750-E. Their major limitation is their stack ring "only" offers dual 16 Gbps ports (advertized as 64 Gbps [dual 16 duplex]).

[Jon Marshall]

Joseph

"With the 6500, since the fabric is "optional", 6500 line cards are designed to work with only the 32 Gbps bus, the 32 Gbps bus and one 8 Gbps fabric connection, dual 8 Gbps fabric, or 1 or 2 connections to the 20 Gbps fabric (provided by the sup720 [720 Gbps] variants)."

I may be reading this wrong but the fabric is not optional with CEF720 cards because they don't support a connection to the shared 32Gbps bus.

Jon

[Joseph W. Doherty]

Jon,

No you're not reading it wrong, I likely wrote it wrong if what I'm trying to describe isn't clear, which is . . .

The fabric is optional to the 6500 chassis, since there's both supervisor and line cards designed to work without it.

When the fabric is provided, some cards won't utilized it, some cards will take advantage of it and some cards won't work without it. For the latter cards, fabric isn't optional for them, i.e. the card, not the chassis.

Of course, you're correct the fabric is standard with the sup720 cards, but it was a separate (optional) card for the sup2 (SFM/SFM2). (NB: if I remember correctly, you could also have two SFM/SFM2 fabric cards to support fabric redundancy.) Also, all 6500 supervisors support the 32 Gbps bus, including the sup720 cards, but not all 6500 supervisors support or provide fabric.

With 4500 series, I believe(?) all the supervisors provide fabric.

[branfarm1]

Thanks guys for the replies. I love the idea of the greater performance on the 6500, as well as the option to use service modules.

What would the symptoms be if I were running into problems with either the backplane bandwidth being insufficient, or port groups being oversubscribed? (I know they are already oversubscribed, but how do I know if it's affecting performance?)

[Joseph W. Doherty]

A reduction of performance or quality between hosts if there's more actual demand than the resources can provide.

PS:

BTW, oversubscription is quite common and often is not too much an issue if properly allocated. If your 100 servers only used their NICs part time (most would), you would likely do alright if you provisioned for the total average plus 50 to 100% if less than the maximum possible.