Nexus F2 Modules clarification

[clark white]

Dears,

Attached are the 2 no's screenshot 1 no's is the screenshot from the datasheet where the cisco  says the he fabric architecture scales beyond 15 terabits per second (Tbps), How cisco calculate figure 15 TB ????

The Nexus 2 screenshot attached  i understood very well and on basis on that screenshot i am calculating but i am not able to reach to the 15 terabits

thanks

[Jon Marshall]

I can't make any sense of the second link you posted :-) but then I haven't had to work out performance figures for Nexus before.

However I can try and help with how Cisco calculate that figure. This is how I understand it -

Firstly the figures is based on an 18 slot chassis so the figures would be different for a 10 slot one.

The 18 slot chassis supports up to 16 I/O modules. So assuming you are running all the ports on the modules as 10Gbps -

48 ports (per module)  x 10Gbps = 480Gbps

16 (number of modules) x 480Gbps = 7680Gbps = 7.5Tbps

Then because they are full duplex -

7.5 x 2 = 15Tbps

Like I say I can't relate this to your second diagram because I'm not sure what it means by traces although if you send me the link I'd be interested to look at it.

Also worth noting that the above calculation obviously does not include the supervisor modules which your second link does.

What is slightly confusing is that for a 10 slot chassis it supports up to 8 I/O modules so you simply divide the above figure by 2 to get 7.5Tbps which again does not include the supervisors.

However your diagram does not come out with that figure. Including the supervisors it comes to 4.14Tbps so something obviously isn't adding up.

I haven't used Nexus switches so the above is simply a calculation based on the datasheets and nothing else. Which means there may well be other factors that I am not aware of which would explain the difference.

Perhaps if you explained how the second one works we could relate it to the official Cisco figures ?

Or someone like Reza may see this and help explain the difference :-)

Jon

[clark white]

Dear Jon,

thank for the reply,

attached are the maths shown from the student guide,

For the F1 fabric modules the bandwidth is 23gbps traces and for the F2 fabric module the bandwidth is 55gbps Trace

[Jon Marshall · Accepted solution]

Okay, after a bit of reading I think I understand where some of the confusion has arisen.

Firstly and most importantly the math performance sheet you posted works out the maximum possible throughput per slot whereas the calculation I provided in my first post is working out the actual maximum throughput.

The difference is dependant on the I/O module. Just because the slot provides a certain amount of throughput this does not mean that there are I/O modules that can use all that throughput ie.

each Fabric-2 module can provide 110Gbps per slot and you can have up to 5 of these modules so you can get a maximum possible throughput of 550Gbps per slot.

The I/O module however can only generate up to 480Gbps which is the actual maximum throughput so there is unused throughput per slot. This is simply because Cisco have not released any I/O modules that can generate more than that (or at least I'm not aware they have).

Which means using your calculation will generate higher numbers than mine.

Which may surprise you because as you said in your first post you were struggling to get to the figure Cisco were quoting.

To get the 15Tbps figure Cisco use the largest chassis (7018) and Fabric-2 modules which means we need to apply your math performance sheet using those as input. So leaving out the supervisors because my initial calculation didn't include them -

16 x I/O modules with dual traces = 32

5 Fabric-2 modules = x5

Total 55Gbps traces  = 160

160 x 55Gbps = 8.6Tbps (rounded up)

Full duplex operation 8.6Tbps x 2 = 17.2Tbps

As you can see this figure is higher than my calculated figure because of the reason explained earlier ie. my figure is dependant on the actual throughput of the I/O module.

Hope all the above made sense.

If it doesn't or you have any more queries please feel free to ask.

Jon

[clark white]

Dear Jon,

Thank for the reply

so i can say that throughput cannot go more than 17.2 Tbps for Nexus 18 slot chassis.

thanks

[Jon Marshall]

For the 7000 series yes I think that is the limit although I haven't read through all the documentation so I can't say for sure.

But the docs I have found suggest this is the limit.

Bear in mind there is now the 7700 chassis so this may be different.

I didn't read any of that documentation as I was too busy on the other stuff :-)

Jon