SMB Cisco design swithces

Unanswered Question
Sep 3rd, 2009


Today our network consist of 3x2960 switches in a flat VLAN domain and it is about 100 users. 50 of these users are heavily accessing a SAN and have 1 Gb access. Does anyone have any suggestion for a design where we want to upgrade to 600 users and 100 of these are heavy SAN "users". I am looking for a access/distribution/core model with Cisco production switch models. Scalability, redundancy and cost is of course an issue. BR, Leif

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 4.7 (3 ratings)
Joseph W. Doherty Fri, 09/04/2009 - 03:41

For 600 users, don't really see the need for a 3 tier network, with current equipment, 2 tier should be fine. However, moving to 600 users you'll probably want to start using routing within your LAN.

Best match for equipment would very much depend on physical topology. For instance, will all links meet at one location or do you intend to have edge equipment out at some distance requiring fiber connections?

One inexpensive option might be 2960G or 2975 switches as edge devices connected to a 3750G stack. Setup two or four port channels between edge switches and the 3750 stack; distribute connections across the 3750 stack members so that individual stack unit failure does not drop the connection to the edge switches. Also connect servers and SAN device to 3750 stack. (NB: you'll be able to route within the 3750s; using the base image.)

lsoleim Sun, 09/06/2009 - 04:51

Thanks for you respons. The physical topology consist of one location today where all the links are met. The servers, routers and switches are located in the basement while the users are deployed in different floor in the building. So I think this is can be described as a swich block. You suggest a 2 tier architecture/collapsed backbone and I fully agree whith this assumption. We are also looking for 10 Gbps througput in the back bone. Does this architecture make this possible with 3750G switches or do we need a separate switch for the server domain where the SAN is located as well? And I appreciate your answer and respons very much. I know that this network is small and simple but we need to get this architecture to comply with the user requirements which is fast file transfer between the SAN and the PC clients. And we also want have scalability in mind since the organization might expand in the future. In the next 6 months a branch with about 50 users will be connected to the main site as well. They will not access the SAN over the WAN but use standard office applications.

BR, Leif

Joseph W. Doherty Sun, 09/06/2009 - 05:53

If cost is important, you'll find 10 gig raises the price quite a bit!

Because of the additional cost, you should consider your backbone bandwidth expected utilization very carefully. You might find two or four gig Etherchannel or discreet edge devices (each with its own uplinks) might satisfy your backbone bandwidth requirement.

For the user edge, there was a 3750 model that supported a single 10 gig port, but it has been discontinued. So that leaves us, for non-chassis switches, some switches that support 10 gig are: 3650-E/3750-E/49x8-10G. (BTW, the 3750-E and non "-E" can be mixed in the same stack, providing such a stack 10 gig uplinks.)

For your core, for non-chassis, you might use 3560E-12D or 4900M. This though leaves you the problem of how to connect your servers. For server switches, and 10 gig uplinks, I would suggest anything but non -E 3750s, similar to user edge.

For a combined core/server device, which avoids bandwidth issues between the server edge and core (and suitable for small deployments), then you'll likely want to condider using a chassis. Assuming you want redundancy, i.e. you'll need two devices. If you want to work with just one logical device, the 6500's VSS is something to consider.

mmacdonald70 Sun, 09/06/2009 - 12:43

I agree. If cost is a factor, I would make sure that 10G is really a requirement for such a small number of users.

Depending on the number of servers that are required, you could go with a pair of 4900M switches with 20 port 10/100/1000 modules. The 3560E-12SD also supports 1000M ports with the use of the twingig module.

Joseph W. Doherty Sun, 09/06/2009 - 16:05

Matthew, excellent addendum concerning the ability to provide other than 10 gig ports on the 4900M and the 3560E-12SD.

Honestly, forgot to mention those capabilities, also for the reason, for the price of this class of switches, seems a waste to not take advanage of the 10 gig ports. Even so, worth highlighting those features.

lsoleim Wed, 09/09/2009 - 04:14

Hi again,

When you say "non -E 3750" you mean 3750 switches?

And I am also a bit confused about how many ports a TwinGig modules does give you. Is it so that if you have 2 X2 Gigabit uplinks you are in fact having four 10 Gigabit uplinks?



Joseph W. Doherty Wed, 09/09/2009 - 04:28

There are two series of 3750s the originals and the later -E versions. Both are current series, but the -E offers higher performance. Also, the two different series can be stacked together.

Non -E:


TwinGig provides two gig ports on a 10 gig module. I.e. either one 10 gig or two 1 gig.

Two X2 modules could provide dual 10 gig, one 10 gig and dual gig, or four gig.

Leo Laohoo Sun, 09/06/2009 - 14:25

The 2350 was released early this year. It's marketed as a SMB server switch. Here are some of the features you may be interested:

• 48 10/100/1000 downlink ports with two 10 Gigabit Ethernet X2 uplink ports,

• Cisco TwinGig Converter Module for migrating uplinks from Gigabit Ethernet to 10 Gigabit Ethernet

• Modular fan and modular AC or DC power supply

• Out-of-band Ethernet management port along with RS-232 console port

Cisco Catalyst 2350 Series Switches

Hope this helps.

Joseph W. Doherty Sun, 09/06/2009 - 16:09

Leo, one series I had overlooked. Nicely fills the less expensive L2 switch option with 10 gig uplinks. Its performance also makes it a very nice candidate as a server edge L2 switch for a small network.

Leo Laohoo Sun, 09/06/2009 - 16:56

Hi Joseph,

Thanks for the rating. It's a new product but not receiving any marketing exposure so I'd be surprised half of the experts (including you) know about this.

I agree with you. Layer 2 with 10Gb is a very attractive option. I'm not sure how much this model cost though. The only downside I have is the lack of StackWise capability (like the 2750 and 3750) and I would like to see this in the future. I guess this is designed for one-switch-per-rack only and you can't really fill up a rack with > 48 ports.

Again, it's a nice option.

Joseph W. Doherty Sun, 09/06/2009 - 17:15

Leo, I almost made a comment on how nice if would be if they stacked. I strongly suspect, reason we don't see some products from Cisco that we might like is they "compete" to closely against other Cisco products. For instance, stackables vs. chassis switches.

Don't know their cost either, but since they're L2 only (as in don't route), I would assume they would cost less than similar L3 switches, i.e. at least those with same PPS and fabric bandwidth.

Leo Laohoo Sun, 09/06/2009 - 17:27

I guess you can "stack" the two using the cluster command.

You'd be hard to find a rack which has a requirement in excess of 48 copper ports for a one-switch-per-rack appliance. This is probably the reason why stacking is not supported.

lsoleim Thu, 09/10/2009 - 12:31

To summarize I have added a physical network diagram based upon the discussions under this thread. I have included the Catalyst 4900M switches for the core layer due to the need for 10 Gigabit access for the about 100 heavy SAN nodes/users. Is it appropriate to divide the VLAN domains based upon the the physical switches?

Leo Laohoo Thu, 09/10/2009 - 14:56

Just out of curiosity but do you want to use a 2960 in your server room? What clients/host connect to it? If clients are ILO/DRAKs then I understand.

Joseph W. Doherty Thu, 09/10/2009 - 17:09

You show a green stack cable between the pair of 4900Ms, but these are not stackable switches. You can link them using their normal ports.

It's unclear the role the 3560Es play in the access layer.

Depending on the need for the gig and 10 gig core ports, a pair of 3560E-12Ds or 3560E-12SDs might work. (Believe they would be less expensive than 4900Ms.)

lsoleim Thu, 09/10/2009 - 22:22

The 3560Es in the access layer are switches for the heavy SAN users. So they have 10 Gig uplinks to Core and SAN. The other 2975 switches are used for administrative/office users. The core need for now 6 10 Gig uplinks and the need for scalability in the future. So it might be a good idea to use the 3560s in the Core as well?

Joseph W. Doherty Fri, 09/11/2009 - 03:34

Re: 3560E-12SDs in access

Depends on what you need for downstream ports, for this model it has the 10 gig uplinks and SFP gig ports. If the gig ports were copper, the 2350 (suggested by Leo) would likely be less expensive (and also provide more gig ports).

Re: 3560E-12Ds in core

Here too depends on number, and type, of ports required. The 3560E-12D only provides twelve 10 gig ports, although these ports also support twin gig. The 4900M, because of its dual modules slots, provides you more port options (and also more ports). However, I would expect this unit to be more expensive vs. the 3560E-12D.

BTW, here's a nice little At-A-Glance whitepaper on using a 4900M in a two-tier collapsed core/distribution, although the 3560E, I believe, could do so as well.


This Discussion