New network build

[psydii]

If you don't have stacking switches or chassis based switches, then switch->switch ->switch->switch is a legitimate arrangement. If you are concerned about bandwidth, you can use 2/4 port LACP trunks between them. With the switches at either end of the chain connected by fibre back up to the core, spanning tree will keep one of the middle switch->switch links down, and you get load balancing across your uplinks.

 

If your comms rooms are sufficiently small then of course you could have an uplink per switch. However two switches with separate single 1GB/s links back to the core will not perform as well as two daisy-chained switches with a single 2 port LACP trunk.

Edited February 26, 2013 by psydii

[januttall]

how do you prevent switching loops? STP cuts off the redundant links so no load balancing can be being done and without STP you make your self a loop?

and you cant say 100% that STP will cut the middle link.

 

I understand the idea but from a redundancy, bandwidth point of view and even a hop's point of view a switch to Many other switches always trumps daisy chaining them. and yes LACP trunks help a lot in any configuration but the data through is always higher if the star is used rather than a buss, your not having as high a contention ratio on each of your switches, E.g 1-48 or 2-48 instead of 2-86 2-144 or even 2-192 and at far fewer hops your reducing the processing time taken to transfer the data packets through each switch as each will want to process the data its self before forwarding.

 

Networking is difficult an no one can realy say they have the best design it depends on requirement situation and cost ultimately.

Edited February 26, 2013 by januttall

[psydii]

Assuming you've got your root bridge configured sensibly, out at the edges spanning tree chooses to block one of the inter-switch ports in the chain, because the best path back to the root bridge is shortest from the two end switches, it leaves in effect two smaller chains attached back to the core.

 

That said, every reasonably sized network I have seen (apart from my current one) employs in effect a matrix of connections between the cabinets with a little daisy-chain for the edge-most switches. In these networks MSTP or EAPS was used to achieve the best throughput and resiliency.

 

 

 

But I agree there is no correct answer, we're just thowing about ideas, what we'd do and what we do do. My current 1Gb/s lan (predominantly a star with quite a bit of daisy-chaining at the edges) cost an order of magnitude less than the BSF one I last looked after. It is simpler, and less resilient, yet operationally I cannot tell the difference, other than I don't need to pay for a CCIE when we want to add a switch.

 

Finally for a different perspective to us lot concerned with small numbers of users.... In some cases it is not always faster for each switch to have it's own single link back to the core: Per Second Measurements Don’t Cut It - Server Fault Blog

Edited February 26, 2013 by psydii

[twin--turbo]

Don't confuse Stacking Switches with Daisy Chaining of switches.

 

A True Stack uses a stack interconnect that offers management of the entire stack and should offer backplane speeds across the entire stack. Daisy chaining require individual management and only port speed between switches.

 

 

I discovered that our supplier had put 8 layer deep daisy chaining with no redundancy on our sister site ( that was within 3 hrs of finally getting the password out of them ) they argued it was acceptable until I sent them a diagram out of Cisco Network assistant highlighting the bottlenecks and multiple points of failure.

 

We run (in reducing quantity),

Core -> Distro -> Edge -> User

Core - > Edge -> User

Core -> user

 

 

Rob

[Gaz]

There are some good ideas and thoughts being thrown about here.

 

Core - > Edge -> User << that was the way I had this planned in my mind, it depends how big the building is going to be I suppose, where they decide to locate the rooms etc etc

 

So do stacked switches use the interconnect to communicate between each other? at what speed does that operate at?

I found this video quite interesting.

[twin--turbo]

So do stacked switches use the interconnect to communicate between each other? at what speed does that operate at?

I found this video quite interesting.

 

Usually you have a stack master , it will tie the others in as a logical unit so you can configure it as one device.

 

the speed is dependant on the brand/model but often is at the full backplane speed of the individual switches so much much higher than 1Gb ports. Good ones you can link the top of the stack to the bottom so that any failure in a member or a cable will not take out more of the stack.

 

Last set I had was 6 years ago though

 

Rob

[psydii]

I'm not confusing stacking with daisy-chaining. Because I've never really been a networking guy, and having looked after procurve, cisco, 3com and extreme as well as having to translate for SLT and governors my language is often confused.

 

7 Deep daisy chain? Owch. I wouldn't want to be the project manager on the receiving end of explanation as to why that was a bad idea.

 

My problem with having a big stack of switches is that in my experience they can be a single point of failure... I have on several occasions had a switch in a stack go bad and take out the connectivity for the whole stack.

 

There are, broadly speaking two types of stacking. The Good type that generally uses dedicated modules and cabling running at 10-40Gbs, in which the switches behave as one, allowing you to have, say, an LLDP group with ports across different switches in the stack. The other type is simply a managment convenience to present one management interface for a number of devices, this commonly uses any 1Gbe port to link the devices. With this latter type there isn't much operational or architectural difference between a stack and a daisy-chain, I suspect it just looks better in a network map.

 

The deepest part of my network is 3 hops with 2Gb/s the whole way to the core. In the event of a failure in that particular cab of either the top or bottom switch RSTP will unblock a port in the middle of the cabinet and the network will become 4 deep. I feel I ought to point out (again) that this network was an order of magnitude cheaper than the MSTP based 10GbE with daual 8800 Procurve Core monster that had a 'better' design. Would I prefer to have A5500 stacks at the edge? Yes, but the current design (and the monitoring of performance) shows that I really don't need them.

 

A final point from me: I would strongly recommend though that you engage with a specialist network company for the design and implementation for layer 1,2 and 3. Make sure they are engaged in the process before architects put pen to paper.

Edited February 26, 2013 by psydii

[twin--turbo]

After getting the password, I got onto the CLI, nothing much seemed to make much logical sense as I worked my way hopping from switch to switch.

 

So I fired up Cisco Network Assistant ( I hardly ever use it ) and mapped out the horror!

 

 

There is some redundancy, but it's not consistent and in some areas not redundant at all.

 

 

 

Rob

Edited February 27, 2013 by twin--turbo

[Gaz]

So is stacking switches (a true stack) akin to RAID then? RAID 5 for example, the stack can suffer 1 loss and still communicate. In the attached image (ph34r my paint skills) is the dashed line (cable) needed for redunancy? for example if switch 2 fails then switch 1 wont be able to communicate with any of the other switches but switches 3-6 will still be able to communicate with each other.

 

Found this which explains it nicely.

http://www-tss.cisco.com/eservice/est-unified/switch/cat2975/remedial/information_files/inf_plan_switch_stack.html

Edited February 27, 2013 by Gaz

[psydii]

It depends on the technology on the switches. That schematic might represent a true stack, or a daisy-chain with STP hopefully keeping one of the ports blocked.

[TheScarfedOne]

That's just what I've been installing...Juniper call it Virtual Chassis. Works nicely

[Gaz]

How many have you linked/stacked together? I'm just trying to work out the topology that may be needed. Do your access switches feed back to the Virtual Chassis? How does the Virtual Chassis handle a switch failure?

[cpjitservices]

Switch Failure ? Virtual Chassis ? VERY rare!!! Its rare you'd get a switch failure on VC, We use Juniper kit all over the UK now and have yet to see a drop out or a failure, we use EX series Switches in VC. If it was to happen the switches would fail over on the next switch so if we had Switch A,B,C,D and D or C failed, all the traffic would fail over to A or B, providing you have your links connected up correctly and configurations are correct of course, with no latency.

 

Even if a switch member within a Virtual Chassis configuration fails, the Virtual Chassis will continue to operate.

[Gaz]

Just the answer I was looking for.

[cpjitservices]

Well if you want VC, go Juniper.

I've come across alot of switch configs in my time and use to run a network with over 100 Nortel switches in a college, but by far and away the EX series switches are the best Ive come across and are pretty cheap aswell.

They are easy to configure via Web Config or CLI, and are built like brick houses!!! We have the EX 4200's. ATM here we arnt using VC admittedly because there's no need but we do elsewhere.

I am also looking to implement Q Fabric at a remote site.

 

Juniper Networks - Virtual Chassis Technology Best Practices - Knowledge Base

Introducing QFabric - It's time to tackle exponential demand

 

We have 2200 series in quite a lot of remote sites, some of them even do alll of the routing between sites with Layer 3.

 

These are what we have:

[Gaz]

I'm not looking at brands just yet but I know Juniper are up there with the best of them. Is a Virtual Chassis just what Juniper called a stacked switch? From reading the Best practices guide I get the feeling it is. Just out of interest what Juniper switch would make a good core switch for an SMB with around 80 computers.

 

And now for a Vlan question

Am I right in thinking that if I've got Layer3 switches at the core and Layer2 switches at the distribution/access points then the Layer3 core would do the inter-vlan routing for the layer2 switches?

[cpjitservices]

We have x2 EX 2200 with 48 ports on each serving a network with 100 nodes. Yes the Layer 3 switch will do the vlan routing for you, thats how we have it at a remote site.

 

VC is upto 10 Juniper EX4500 switches, its basically an extra cable you buy with the switches.

 

Depending on sites, usually anything with 200 or less nodes we use EX2200 series switches.

 

So for 80 users, 2 EX2200's (POE if needed) will be perfect.

Edited March 1, 2013 by cpjitservices

[john]

Another happy Juniper user here I have 4 EX4200s (should become 5 in the very near future....) in a VC stack which is split between two campuses 2 on each and they are linked by 10Gb fibre so even shiner

 

The EX2200s are great, we had one that had a borked firmware due to power spike and it got a bit stressed and carried on routing traffic for existing connections which meant I didn't actually notice the issue till I couldn't get into it to tweak a VLAN setting.