Switching Capacity Query

[Steve21]

Evening Guys,

 

Wonder if someone could explain this bit as I'm not sure if I'm missing an obvious or whether resellers just trying to make me upgrade for no reason

 

If a switch has for example - Throughput : 65.4 Mpps ¦ Switching capacity : 88 Gbps I'm assuming there's absolutely no reason that this can't do full Level 3 routing of up to 88 Gbps.

 

So for example if we have 2x 10G fibres going into it, and then a few servers say 20x 1G connected. So total switching if both in and out maxed = 2x 2x 10G and say 20x 2x 1G = 80G total switching if everything is at max speed?

 

Whether there's 10 or 10000 computers coming into it over the fibre it shouldn't affect the outcome of the switch as it'd be maxed by the 10G connections in rather than the switching right? There's no hidden formula like VLAN routing = 2x capacity or something stupid

 

Thanks,

Steve

[FN-GM]

Its worth noting that routing will be using your switches processor. Switching uses ASICS, a physical device one for each port, its on a per port basis. With ASICS it works at wire speed, so if every port is connected and running at full speed, you will have enough capacity. Each port is guaranteed to have enough capacity to run everything at full speed, you can't max out the switch by fully populating each port at full speed, they are not a shared resource. However the switch might not have the processor to do the routing and other functions such as VRRP, OSPF etc that are not specific to a port. (This is all assuming you use something like Cisco or HP and not cheap makes).

 

In the Cisco world you use SDM templates to allocate your resources, so one thing cannot hog all the resources and bring the switch down. On the switch using the show commands you will be able to see what capacity you have and figures on how it can be used.

 

FYI: If you have a router not a switch this uses the processor not ASICS. That is why routers have more features because its all software based and not reliant on ASICS. However this is where you can get performance issues.

 

Those figures they are presenting might be just to total processing power of all the ASICS combined. If so, for routing it might be then be useless.

Edited January 24, 2016 by FN-GM

[jtotheb]

If a switch has for example - Throughput : 65.4 Mpps ¦ Switching capacity : 88 Gbps I'm assuming there's absolutely no reason that this can't do full Level 3 routing of up to 88 Gbps.

 

Keep in mind that's no doubt max performance going downhill with a good push. If it can packet switch in hardware then it can possibly get close to it.

 

Whether there's 10 or 10000 computers coming into it over the fibre it shouldn't affect the outcome of the switch as it'd be maxed by the 10G connections in rather than the switching right?

 

With insane numbers you may approach the limit of the device's MAC address table!

 

There's no hidden formula like VLAN routing = 2x capacity or something stupid

 

No there's not, but the routing mechanism the switch uses will determine its performance.

 

Its worth noting that routing will be using your switches processor.

 

A good modern L3 switch is going to route in hardware at wirespeed. It's only going to process switch if something's gone wrong or the other mechanism's been disabled (e.g. CEF on a Cisco switch which isn't particularly CPU intensive).

 

With ASICS it works at wire speed, so if every port is connected and running at full speed, you will have enough capacity. Each port is guaranteed to have enough capacity to run everything at full speed, you can't max out the switch by fully populating each port at full speed, they are not a shared resource.

 

That's dependent on the model of switch. Even older Cisco 3750 models I've still got in use couldn't run every port at full whack simultaneously as the underlying switch fabric can't support it.

[Steve21]

Thanks guys.

 

Aye guess my concern is just the fact that one of our resellers is basically insisting that we'll need a 10G switch to cope the routing (600-1000 devices) e.g. a Dell N4000 stack even if we're not using 10G connections, and it seems overkill spending £10k+ on a switch if it's not needed. (Before considering issues with SFPs vs CAT etc)

 

Even one of the baby 5800s does 200Gb+ at 3k~ each and seemed to make more sense saving the huge difference to spend elsewhere (There's no plans for 10Gb connectivity in the next few years cycle)

 

Doesn't help that everyone you talk to has a different opinion, Dell for example suggest an even higher switch

 

Steve

[jtotheb]

This is as a core switch?

 

What about something modular like a Cisco Catalyst 4500 series? If you ever go to 10 gig you buy the (hilariously expensive) 10 gig line card(s) when you need it.

[njc235]

I know I always bang on about them but have you considered HP's virtual chassis technology? That would allow you to deploy a mix of 10G and 1G now and add more 1u members later to offer more of either. Look at the 5800 and 5900 ranges and all of their options.

[Steve21]

Aye been looking at the chassis options (albeit the HP as just easier to spec online I find) Only concern is they are a single point of failure still (although a small one if everything's doubled up)

 

And yep been looking at the 5800/5900s, but again only concern is if I am splitting them between two rooms e.g. one production and one DR, without dual homing every switch uplink it's another point of failure. Unless you have 2 each side?

 

Maybe I'm worrying about single failure too much, just seems a shame not to have the options if I am replacing stuff

 

Thanks,

Steve

[IanT]

This is as a core switch?

 

What about something modular like a Cisco Catalyst 4500 series? If you ever go to 10 gig you buy the (hilariously expensive) 10 gig line card(s) when you need it.

 

Cisco Catalyst 4500 Series is discontinued isnt it?!

[jtotheb]

Cisco Catalyst 4500 Series is discontinued isnt it?!

 

Various supervisors, chassis (is chassis the plural of chassis?!) and software options are EOL, but the 4500E with an 8E supervisor card running 15.2+ is still current afaik. However, Cisco seem to be pushing the 4500 as a high density access switch with the 6500/6800 positioned in the core portfolio.

[bossman]

@Steve21

 

Have look at these, they come highly recommended and you can get them for a reasonable price from a good third party supplier:

 

https://meraki.cisco.com/products/switches#models

 

Its what I am looking into atm

[Steve21]

How much are those bad boys looking to set you back?

 

Steve

[cpjitservices]

Juniper... EX Series.

 

Marginally cheaper than Cisco and in my opinion a better overall device.

[bossman]

@Steve21 @cpjitservices

 

Juniper are good although there has been some code irregularities regarding patching which has left some devices open to external intrusions, although Meraki have been bought out by Cisco the pricing is very good for the level of equipment actually cheaper than Cisco switches, the software is excellent for the Meraki switches too

 

Just looking atm and gathering costs for feasibility report,

[cpjitservices]

agreed @bossman but the patch was almost instant!! , I dont think it affected the switches is was the SX Series firewalls I believe that where affected.

[bossman]

@cpjitservices

 

I think if I remember? It was mainstay routers which were badly affected as well as enterprise switches but as you say they had a patch out immediately but that doesn't detract the severity of the software fault,

 

Hope you are well my good friend