Chassis switches vs stacked switches

FRAGMAN BOB

Limp Gawd
Joined
Jan 25, 2005
Messages
292
Is there an advantage to having chassis switches such as a 4510, vs having stacked switches?

We are looking at a closet that will need ~500 patch connections. Our Cisco partner recommends two 4510R switches. We currently have two older 3COM 7700. Our biggest qualm with the chassis switches is the wire management. There is no possibility to use something like a neatpatch when the 48 port blades are filled up.


Is this just how it is for high-density closest or is there a better design for this?
 
It comes down to price, most of the time. If you're doing more than, say, 6-7 switches in a stack, I'll typically go with a chassis instead. You get better performance/throughput out of a chassis in most cases.You also tend to get many more features out of a chassis switch, which may or may not be useful, depending on the deployment.
 
Chassis offer several advantages,

1) only need 1 set of redundant PSU rather than (2) psu per 1u switch. reduces cabling
2) if you're only physically stacking (not a logical stack with cables) the amount of management interfaces needed is greatly decreased. this is why we have chassis in pretty much any closet that needs >48 ports.
3) cabling is generally easier due to not having to deal with uplink ports or switch stacking cables.
4) chassis are usually designed for a higher mtbf, with hot swappable cards & fan trays, and in some chassis redundant management cards
5) usually the backplane on a chassis is going to be much quicker.

With vertical wire management its very easy to manage a fully loaded chassis if you use common sense, velcro, and cables that aren't of excessive length.

I am a huge fan of chassis switch where they are needed. Not every closet necessitates a chassis.
 
I wish I had been around when we ordered the 7 switches in our main panel. I would have gotten a chassis setup. It is a mess as it is, but it works.
 
We went with stacked 3750x's in a couple of data closets recently. The most I've done in a stack so far is six. Our biggest reason with going with stacked 3570x's was it would be easier for us to keep replacements on hand. I can't wait 4 hours for a part replacement and can keep everything I need to replace a 3750x on a shelf at a more reasonable cost.

Granted the chassis gives you some other benefits as listed above, but for us it made more sense to pull out the 4000 series chassis and replace them with stacked switches. 3750x is also a high end switch. In fact I think I could argue that all 5 of green's points are moot with the x series switch.

edit: So the 4510 is going to give you an increase of 150k on your MTBF over a 3750x 48 port. My question for everyone is whether that is the entire chassis or a blade. If it is an entire chassis I think I'd still rather deal with a single switch failure (48 users) than all 6 blades failing(288 users). If I'm wrong here please correct me, I think this is an interesting debate!
 
Last edited:
We went with stacked 3750x's in a couple of data closets recently. The most I've done in a stack so far is six. Our biggest reason with going with stacked 3570x's was it would be easier for us to keep replacements on hand. I can't wait 4 hours for a part replacement and can keep everything I need to replace a 3750x on a shelf at a more reasonable cost.

Granted the chassis gives you some other benefits as listed above, but for us it made more sense to pull out the 4000 series chassis and replace them with stacked switches. 3750x is also a high end switch. In fact I think I could argue that all 5 of green's points are moot with the x series switch.

edit: So the 4510 is going to give you an increase of 150k on your MTBF over a 3750x 48 port. My question for everyone is whether that is the entire chassis or a blade. If it is an entire chassis I think I'd still rather deal with a single switch failure (48 users) than all 6 blades failing(288 users). If I'm wrong here please correct me, I think this is an interesting debate!

That reasoning doesn't make a lot of sense. Many copper blades for a chassis are going to be cheaper than a 3750. It'd be more cost effective (in many/most cases) to keep some spare blades on hand than spare switches. Replacing a blade is usually a hell of a lot easier than replacing a switch too.
 
Wow great responses. Thats what I figured with the chassis switches being superior in most ways.

I'll watch that panduit video, hopefully get some good ideas.
 
That reasoning doesn't make a lot of sense. Many copper blades for a chassis are going to be cheaper than a 3750. It'd be more cost effective (in many/most cases) to keep some spare blades on hand than spare switches. Replacing a blade is usually a hell of a lot easier than replacing a switch too.

It depends on the business case, in my opinion. I've got 40 closets with 3750x stacks in them. Only two of them have more then 4 switches (which would be the point I would consider a chassis). Each closet has a minimum of two switches. So, to keep it easy, let's say there are 90 3750x's deployed. If I replaced 10 of those with 2 4500 chassis I would still have 80 deployed.

I'm going to say that an ethernet blade is about $10k, same as a fully configured 3750x. I don't know the actual prices, but it has to be close to that. That said, I can put two blades on a shelf to have in case of emergency, but those blades will only be useable in two closets.

On the other hand, I can buy two 3750x's and have them be useable in all 40 closets. For us, it makes more sense to have something sitting on a shelf that we can use in any closet. Even if the copper blade is half as much I still can't use it in every closet so it's cost on a shelf is actually greater.
 
It depends on the business case, in my opinion. I've got 40 closets with 3750x stacks in them. Only two of them have more then 4 switches (which would be the point I would consider a chassis). Each closet has a minimum of two switches. So, to keep it easy, let's say there are 90 3750x's deployed. If I replaced 10 of those with 2 4500 chassis I would still have 80 deployed.

I'm going to say that an ethernet blade is about $10k, same as a fully configured 3750x. I don't know the actual prices, but it has to be close to that. That said, I can put two blades on a shelf to have in case of emergency, but those blades will only be useable in two closets.

On the other hand, I can buy two 3750x's and have them be useable in all 40 closets. For us, it makes more sense to have something sitting on a shelf that we can use in any closet. Even if the copper blade is half as much I still can't use it in every closet so it's cost on a shelf is actually greater.

Deploying chassis switches in your case wouldn't be cost effective as you don't have the density to justify it. I assumed from your post that you have much more density. With less than four switches per closet, it makes no sense to run chassis, IMO.

To clarify though, the WS-4648-RJ45V+E (the blade I'd use in most closets) is only $7500 list. That meant ~$4k cost to customer. The WS-3750X-48PF-L (LAN base/barebones full PoE 3750) is $11,400. That's ~$6k cost to customer. Pretty big difference when you're talking about keeping cold spares around, and that switch is the lowest license 3750X there is, which means no stack power and less features. In the case of the 4500, the blades are typically going to be cheaper than a 3750X, especially if you need advanced features.
 
One feature that many chassis switches support are the ability to set the MTU per port as oppossed switch wide... for instance if you have clients that need 1500 byte mtu because 9014 causes errors you can just set it for those ports and for server that need jumbo or baby giants you can set mtu for those particular ports.

Atleast in the Cisco world ... my 3750E cant set per port MTU but a 6500 series chassis can.
 
Deploying chassis switches in your case wouldn't be cost effective as you don't have the density to justify it. I assumed from your post that you have much more density. With less than four switches per closet, it makes no sense to run chassis, IMO.

To clarify though, the WS-4648-RJ45V+E (the blade I'd use in most closets) is only $7500 list. That meant ~$4k cost to customer. The WS-3750X-48PF-L (LAN base/barebones full PoE 3750) is $11,400. That's ~$6k cost to customer. Pretty big difference when you're talking about keeping cold spares around, and that switch is the lowest license 3750X there is, which means no stack power and less features. In the case of the 4500, the blades are typically going to be cheaper than a 3750X, especially if you need advanced features.

you're leaving out the overall cost of the chassis and the sup's in the calculations. my guess is it would be roughly equal if you did the math and averaged the TOTAL(chassis, sup, and blade) between a newer 3750x stack and a chassis.
 
I'm guessing the follow-up to this discussion is how do you make your chassis switches cabling look pretty? I've looked in the networking gallery but most of the chassis switches there only had a blade or 2 in them. I'm talking a 7-10-blade switch that is able to have a a cable easily swapped.
 
the youtube video posted by timberdoodle gives you a very good starting point. i wouldnt take what they're offering as the gold standard, but it's a great foundation. my own personal recommendation though - DO NOT route cables over the fan tray and/or power supplies. they WILL fail, or need upgrades or something will happen and need replaced/swapped.
 
I'm guessing the follow-up to this discussion is how do you make your chassis switches cabling look pretty? I've looked in the networking gallery but most of the chassis switches there only had a blade or 2 in them. I'm talking a 7-10-blade switch that is able to have a a cable easily swapped.

Good luck. Our 6509 cores have 2 fiber connections to every data closet and I don't know what I'll do when one of the blades fail. I'm interested to hear what others are doing though.

We split it down the middle and use a ton of velcro, it'll still be a nightmare. The best thing I can offer is to make sure you have everything labeled and documented. The worst case in the event of a failure of any kind is being unprepared to handle it.
 
you're leaving out the overall cost of the chassis and the sup's in the calculations. my guess is it would be roughly equal if you did the math and averaged the TOTAL(chassis, sup, and blade) between a newer 3750x stack and a chassis.

I'm not though. We're talking about cost to keep hardware on hand. I've already conceded that a chassis doesn't make sense in his case. It does, however, make sense when you have 6-7 3750s in a stack. It becomes cheaper to use a chassis at a certain point, which depends on bundles and other factors.
 
I'm not though. We're talking about cost to keep hardware on hand. I've already conceded that a chassis doesn't make sense in his case. It does, however, make sense when you have 6-7 3750s in a stack. It becomes cheaper to use a chassis at a certain point, which depends on bundles and other factors.

I do agree with this. I just think everyone's business case is different. Our biggest stack has 6 switches in it and sometimes I wish we would have went with a chassis. On the other hand, the 3750x platform is so modular and easy to work with it made sense to follow the standard we already had in place.

Like Vito said - in a dense deployment it makes more sense to go with a chassis. If we had 5 or more data closets that were that dense you can bet we'd go with a chassis. However we purposely keep low port densities because I work for a hospital system. Each floor has at least two data closets to prevent a closet failure from taking an entire floor down.
 
Back
Top