Access and Distribution/Core Switches

Eildor

I want access layer switches with 48 x Fast Ethernet ports and 4 x Gigabit Ethernet ports.

I want distribution/core switches with 48 x Gigabit Ethernet ports and 4 x 10 Gigabit Ethernet ports.

I have had a look at the Cisco products page however the sheer number of models and lack of clear port speed information (unless I'm looking in the wrong place) has left me confused. This is for a University project where I will be designing a network that can support 250 users, so considering it's quite a small implementation I don't think I would need the latest and greatest switches. I wonder if there is some sort of tool which allows you to input your requirements and filter all Cisco products, because that would be great.

Thank you

networker050184

Well, isn't the point of the University project to do the research yourself?

vinbuck

Is there requirement you have to use Cisco? If you have a limited budget, I would be looking at other vendors or maybe a vendor mix. Most networks are not pure Cisco these days.

Eildor

vinbuck wrote: »

Is there requirement you have to use Cisco? If you have a limited budget, I would be looking at other vendors or maybe a vendor mix. Most networks are not pure Cisco these days.

I can do whatever I want to be honest, the entire scenario and requirements are entirely up to me... but as I will be writing the configurations I am sort of limited to Cisco as I haven't touched non-Cisco hardware and therefore would have to learn an entirely new vendors CLI syntax.

Eildor

networker050184 wrote: »

Well, isn't the point of the University project to do the research yourself?

This project is supposed to go beyond what I have learnt at University, therefore I am not expected to know everything and it is expected that I will be researching certain topics. I would consider asking questions on a public forum to be a form of research, I'm sure you have used search engines and forums when you've had to research something, I don't see a great deal of difference. I have had a look at the Cisco website, however I didn't come across a switch which meets the requirements I have specified at a reasonable price, so I thought I would ask here as I'm sure you guys would be able to help.

afcyung

Try Ebay for cheap gear. You might be able to get the stuff you want for the price you can afford. As far as what model does what you need, I have no idea.

Eildor

afcyung wrote: »

Try Ebay for cheap gear. You might be able to get the stuff you want for the price you can afford. As far as what model does what you need, I have no idea.

I wont be buying any of the hardware, but I do need to have an idea as to how much everything costs to make this at least a bit realistic.

it_consultant

Eildor wrote: »

I want access layer switches with 48 x Fast Ethernet ports and 4 x Gigabit Ethernet ports.

I want distribution/core switches with 48 x Gigabit Ethernet ports and 4 x 10 Gigabit Ethernet ports.

I have had a look at the Cisco products page however the sheer number of models and lack of clear port speed information (unless I'm looking in the wrong place) has left me confused. This is for a University project where I will be designing a network that can support 250 users, so considering it's quite a small implementation I don't think I would need the latest and greatest switches. I wonder if there is some sort of tool which allows you to input your requirements and filter all Cisco products, because that would be great.

Thank you

This is what I use:

HP Networking Switch selector tool

You can sort by port speed, moduarity, POE/non-POE, fully managed, web managed, etc. HP is the second largest switch producer behind Cisco.

For 250 users I would buy one 5400 series HP chassis with the 10 GB modules and then 24 port blades (I think you can only get those in 1GB form nowadays)for your endpoint access needs. One box, core and distribution on the same backplane - problem solved on a good budget with high quality hardware. For extra points I would outfit it with 4 power supplies and put the power supplies into separate circuits in your datacenter.

Pricing is hard to come by on HP too so I would take the part number and put it into Amazon and tack on 30%. I would NEVER buy HP equipment on Amazon but I assume you don't have access to Ingram Micro or one of the large resellers. I suppose you could use CDW too...

Eildor

it_consultant wrote: »

This is what I use:

HP Networking Switch selector tool

You can sort by port speed, moduarity, POE/non-POE, fully managed, web managed, etc. HP is the second largest switch producer behind Cisco.

For 250 users I would buy one 5400 series HP chassis with the 10 GB modules and then get 2x 24 port access ports (I think you can only get those in 1GB form nowadays). One box, core and distribution on the same backplane.

Yeah, man, I had a look at that earlier today and it's a shame Cisco doesn't have something similar to make it easier to filter through their products. I think in such a small network it might actually be better to go with HP, but like I mentioned previously I have never touched a HP router/switch and so I'm concerned as to how difficult it will be to get used to the CLI... and I don't even know if there is an equivalent to Packet Tracer/GNS3 for HP routers/switches.

ptilsen

Eildor wrote: »

I can do whatever I want to be honest, the entire scenario and requirements are entirely up to me... but as I will be writing the configurations I am sort of limited to Cisco as I haven't touched non-Cisco hardware and therefore would have to learn an entirely new vendors CLI syntax.

I'm going to 2nd recommendations for HP Procurve, and I can tell you the syntax is pretty comparable. I studied and practiced on Cisco and had IOS and CatOS syntax down fairly well, and when I started working on Procurves it translated very well. Commands are usually roughly the same, but you have to get used to the HP way of doing things.

nerdydad

3560's for your access layer switches, 3750 for your distribution, or collapsed core distribution.

I saw one of your other posts, pm me if you want me to look over your diagrams or would like further clarification, I can get you the exact model numbers.

DevilWAH

depends what you need at the access layer and core/distribution layer.

you could use cisco 3550's for the access and 3560's for the core quite happily for some networks with the base IP image on them. but for othe networks with more complex requirements you would need different feature sets.

This is what you need to look at more than the hardware model. For basic L2 switching a 3550 will preform at much the same rate as a 3750. ITs only when you start needing other features such as routing redundancy, prvt Vlans, routing, etc, taht the different models and different feature sets come in to play.

For example I had a replica of a 1500 user network, that ran on 4500 ands 3750s. using a few 3750s and a stack of 3550's for testing. And it gave me 95% the functions of the real thing although much less utilisation that the real thing of course. (a difference in cost of about £250K)

you need to figure out what your network needs to support in terms of functions and work out what feature sets support it (Cisco or other wise), then look at what hard wear supports these features. for example if you need ospf or Eigrp routing you will need the advanced feature set. and this is a different cost to the standard feature set.

it_consultant

Eildor wrote: »

Yeah, man, I had a look at that earlier today and it's a shame Cisco doesn't have something similar to make it easier to filter through their products. I think in such a small network it might actually be better to go with HP, but like I mentioned previously I have never touched a HP router/switch and so I'm concerned as to how difficult it will be to get used to the CLI... and I don't even know if there is an equivalent to Packet Tracer/GNS3 for HP routers/switches.

You will find most networking equipment has a similar CLI to Cisco. I just configured two ad tran routers the other day that might as well have been Ciscos. You have to understand why Ciscos are so hard to buy and to configure online, their business model is to funnel the selling of their devices through resellers. HP is very similar to this, which is why their switch selector is somewhat hidden.

nerdydad

DevilWAH wrote: »

depends what you need at the access layer and core/distribution layer.

you could use cisco 3550's for the access and 3560's for the core quite happily for some networks with the base IP image on them. but for othe networks with more complex requirements you would need different feature sets.

Yeah, my current mindset is that we only use devices that are not EOL, pretty much as soon as the EOL date is announced, we can no longer use it in a new design, it's thier way of ensuring things stay up to date.

We do use HP switches at the access layer on occasion, but because it is an EIGRP network, almost everything is Cisco.

DevilWAH

nerdydad wrote: »

Yeah, my current mindset is that we only use devices that are not EOL, pretty much as soon as the EOL date is announced, we can no longer use it in a new design, it's thier way of ensuring things stay up to date.

We do use HP switches at the access layer on occasion, but because it is an EIGRP network, almost everything is Cisco.

Indeed, best practice would suggest not going near End of Life kit, which is some thing else to take in to consideration when choosing he hard ware. You see a lot of people asking for help suggesting equipment to buy.

Enterprise equipment is juts like home equipment, you can go in to a computer shop and buy a broadband router for £25 or the one next to it for £150. Both will say 54mbps, and to the casual eye will look identical in all but price. And for 90% of homes uses the difference is not noticeable, but that does not mean they are the same, or that every one should be the £25 because it is cheaper. To make an informed Choice you need to know what it will be used for and if it will support the features you require.

Out of interested a senior engineer was saying the other day that CISCO is releasing EIGRP to non Cisco devices. Any one else hard any thing about this? That is would become a no-proptiory standard?

lrb

Don't just assume that Cisco gear is going to always be the best based on how big you perceive them to be in the networking game. If it's a university project, use this opportunity to see what else is "out there" in the market by asking questions and reading data sheets on vendor websites.

Anyway, moreover to your original question: A recent customer we used Juniper EX4200-48T (48 port 1Gb copper switches) for the access switches and four stacked Juniper EX4200-24F (24 port 1Gb SFP switches) for the distribution switches. Theres plenty of part numbers and stuff on juniper.net if you need these for your university project doco.

Forsaken_GA

DevilWAH wrote: »

Out of interested a senior engineer was saying the other day that CISCO is releasing EIGRP to non Cisco devices. Any one else hard any thing about this? That is would become a no-proptiory standard?

I hadn't heard anything about that, but it would be a good move.

In order to become a standard, it would have to go through a standards body like the ietf or the ieee, and they're not likely to leave the protocol as it is, standard bodies like to tinker. Cisco traditionally goes first to market with a proprietary protocol and then implements the standards when they're approved. This is why Cisco supports both HSRP and VRRP, for example.

More likely than not, Cisco would just release the protocol into the public domain and give a carte blanche license to use it.

Forsaken_GA

lrb wrote: »

Don't just assume that Cisco gear is going to always be the best based on how big you perceive them to be in the networking game. If it's a university project, use this opportunity to see what else is "out there" in the market by asking questions and reading data sheets on vendor websites.

Well, I've never been a proponent of the Cisco answer is always the right answer, but you can never ignore Cisco. They're still the big gorilla, even though they might be 8 tons now instead of 10 tons. Cisco basically created the enterprise and SP network kit markets, and they've basically done a crapton of the R&D for the protocols we use these days. This is why that, despite the fact that Cisco certs are not vendor neutral, they are respected throughout the industry. After all, OSPF is OSPF, and configuring it on a Juniper as opposed to a Cisco router is a matter of syntax, it operates with the same rules no matter what gear is in the mix.

Cisco's made some bad moves in the past few years, and it has cost them market share, but they are still quite, quite relevant, and I doubt you'll find anyone who's ever been fired for recommending a purchase of Cisco gear

unclerico

One other thing you might consider is that a lot of the vendors are promoting the idea of flattening the network and getting away from the traditional three-tier design. Most of them have fancy names for this config, but all it really is is extending the stacking capabilities of the switches over the 10/40 gig interfaces to the closets. This way you technically have one large distributed fabric that is managed as one device. Most all of the big players have this offering.

vinbuck

Add another vote for Adtran...config is almost identical to Cisco and it's as stable as a Cisco box but a whole lot cheaper. We use both provider and enterprise Adtran gear and it plays well with our Cisco MPLS core most days.

Forsaken_GA

unclerico wrote: »

One other thing you might consider is that a lot of the vendors are promoting the idea of flattening the network and getting away from the traditional three-tier design. Most of them have fancy names for this config, but all it really is is extending the stacking capabilities of the switches over the 10/40 gig interfaces to the closets. This way you technically have one large distributed fabric that is managed as one device. Most all of the big players have this offering.

Well the other thing driving flatter networks is virtualization. Those pesky layer 3 boundaries are a bit of a problem when you're trying to move VM's between hosts, so building big layer 2 domains is again in fashion in the datacenter world.

Eildor

Thank you all very much for your input, I REALLY appreciate it. I have had a think about the project and I have decided that I want to keep it to pure Cisco gear as I want to use this project as an opportunity to put everything I have learnt over the past year and a half or so into practice.

Again, thank you all very much for your help!

E-mail coming your way nerdydad!

it_consultant

Forsaken_GA wrote: »

Well the other thing driving flatter networks is virtualization. Those pesky layer 3 boundaries are a bit of a problem when you're trying to move VM's between hosts, so building big layer 2 domains is again in fashion in the datacenter world.

I avoid routers like they are the plague. Switch if you can, route if you have too.

Turgon

Forsaken_GA wrote: »

Well the other thing driving flatter networks is virtualization. Those pesky layer 3 boundaries are a bit of a problem when you're trying to move VM's between hosts, so building big layer 2 domains is again in fashion in the datacenter world.

Check out OTV

vinbuck

Forsaken_GA wrote: »

Well the other thing driving flatter networks is virtualization. Those pesky layer 3 boundaries are a bit of a problem when you're trying to move VM's between hosts, so building big layer 2 domains is again in fashion in the datacenter world.

I thought the Data Centers were embracing MPLS in a big way and using EoMPLS/VPLS to satisfy their need for Layer 2 across Layer 3? Data Centers aren't really my area of expertise as an SP Engineer, but I have read in several places about increased use of MPLS in the data center.

vinbuck

it_consultant wrote: »

I avoid routers like they are the plague. Switch if you can, route if you have too.

It works for some networks, but I would much rather leave load balancing in the realm of a Layer 3 protocol like OSPF and have full bandwidth usage than mess with splitting my VLANs up between redundant links in Multiple Spanning Tree to get sorta the same result. It just depends on what your hardware can do - if running Distributed CEF then I would route where I can and switch when I have to because it's gonna run close to the same speed.

Some designs can do it and some can't but given the choice, I would probably utilize Layer-3 switching over Layer-2 if possible.

it_consultant

vinbuck wrote: »

I thought the Data Centers were embracing MPLS in a big way and using EoMPLS/VPLS to satisfy their need for Layer 2 across Layer 3? Data Centers aren't really my area of expertise as an SP Engineer, but I have read in several places about increased use of MPLS in the data center.

My datacenters are all VPLS based. Its the only way to practically deliver multiple layer 2 connections over one piece of fiber/copper. The logical architecture is still layer 2 since my ESX hosts can see each other as if they were both plugged into a dumb switch. There is a definite demand for easy to use layer 2 connections. I would know too, I demand them!

it_consultant

vinbuck wrote: »

It works for some networks, but I would much rather leave load balancing in the realm of a Layer 3 protocol like OSPF and have full bandwidth usage than mess with splitting my VLANs up between redundant links in Multiple Spanning Tree to get sorta the same result. It just depends on what your hardware can do - if running Distributed CEF then I would route where I can and switch when I have to because it's gonna run close to the same speed.

Some designs can do it and some can't but given the choice, I would probably utilize Layer-3 switching over Layer-2 if possible.

You are looking at this from a different perspective than I am. In a campus LAN environment there aren't that many reasons to have layer 3 separation and when you run into them, they are normally for things like guest and wireless guest networks and telephony. Even in that scenario we are only looking at a handful of routes of which, most traffic will never traverse that barrier - and it shouldn't.

Not to break into semantics, but isn't layer 3 switching...routing?

vinbuck

it_consultant wrote: »

You are looking at this from a different perspective than I am. In a campus LAN environment there aren't that many reasons to have layer 3 separation and when you run into them, they are normally for things like guest and wireless guest networks and telephony. Even in that scenario we are only looking at a handful of routes of which, most traffic will never traverse that barrier - and it shouldn't.

Not to break into semantics, but isn't layer 3 switching...routing?

Routing and Layer-3 switching accomplish the same thing but are different tasks within a router

Most routers do layer-3 switching these days (this is what CEF does) meaning the entire routing table (Routing Information Base - Control Plane) is compiled and placed into the Forwarding Information Base (Data Plane) which then builds a map of which ports belong to a certain subnet so that when a packet comes in, it is Layer-3 switched between ports.

True routing means that the router must consult the routing table before forwarding a packet (certain types of traffic still get punted to the processor, but it's the exception rather than the rule) which involves the processor and is very slow and inefficient. When you read literature that talks about routing at wire speed, they are talking about Layer-3 switching using some form of CEF. The terms are still used pretty interchangeably, but when you read Cisco's docs on it, they make a clear distinction between the two.

It's easier to see the separation on equipment that runs redundant Route/Switch Processors, because if you have a failover and Stateful Switch-Over occurs, the routing table could be unavailable for a few milliseconds or even several seconds, but traffic will continue to flow because the routing table isn't how the actual packets are forwarded at the hardware level. That's what the FIB does.

TCAM (Ternary Content-Addressable Memory) is how this happens at the hardware layer

https://supportforums.cisco.com/docs/DOC-15833

Here is a great link that explains Multi-Layer switching better than I can

https://learningnetwork.cisco.com/thread/21889

ITtech2010

I would definitely recommend 3550/3560 switches as access and distribution switches and maybe 3750 as the core.

Forsaken_GA

vinbuck wrote: »

I thought the Data Centers were embracing MPLS in a big way and using EoMPLS/VPLS to satisfy their need for Layer 2 across Layer 3? Data Centers aren't really my area of expertise as an SP Engineer, but I have read in several places about increased use of MPLS in the data center.

Check out OTV

These are, when you get down to it, hacks, to have to avoid an entire redesign. For awhile there was a movement to break up the huge layer 2 domains and move to a layer 3 routed model, since chomping your bandwidth in half thanks to STP blocking links was getting painful.

However, the requirements for layer 2 adjacency changed the entire paradigm, and now we employ a whole lot of new tech to make our networks appear to be flat layer 2 domains again. I find it quite amusing that we're developing and deploying new tech to take us back to where we were before.

And for the record, I'm with Vin. I much prefer to handle things like load balancing and my qos and such at layer 3, its a much bigger pain in the ass at layer 2, and the bottom line is that with a layer 2 domain, I end up having to sacrifice capacity for redundancy, which is something I don't have to do at layer 3.

But then, you can't move VM's around.

Figuring out creative solutions to this crap is why they pay folks like all of us the big bucks.