Switching capacity and Switch Fabric performance.

[Guillaume]

Hi guys,

I wonder if someone can explain me the difference between them.

My knowledge about the Switching capacity is that if we have a 48 Gig port switch, the switching capacity should at least be 48Gbps in the case of all port being used at full speed. It's the capacity to deal with all the bandwind in same time. Some switch under the 48Gbps capacity for a Gig switch are considered as blocking.

So if we take a stack of 8x4800 series, it says 384Gbps in fact sheet(8x48Gbps).

Where im lost is when a stack of 8x5500 series can go up to 640Gbps. Why the stack would need that much capacity if all the port together can only ask for 384 Gbps ?

Also, what's the meaning of Switch Fabric performance ?

And at last, if someone can tell me the switching capacity of a single 4550T, it's hard to find they always use the 4548GT to represent 4500 series. I found 320 Gbps for 8x4548GT. BTW it mean that those 4548GT can be considered as blocking ?

Thanks for your help !

[Telair]

I see a small miscalculation you made.  1Gig ports support 1 Gb/sec full-duplex.  So both ways.  That means each port can generate 2Gb/sec of traffic.  That means for a 48 port Gig switch it needs to be able to do a minimum of 96Gb/sec in order not to be blocking if in the unlikely event all the ports are running at 100% capacity.

I didn't find anything for the 4550T directly, but for the 4524's, 4548's and 4526 switches I found these stats.

Switch Fabric performance: 48.8Gbps to 184Gbps
Frame forwarding rate: 6.6 to 72Mpps
Stack Throughput: 384Gbps

I would think a 4550T should fall in this range.

[Guillaume]

Wow... It's actually a big miscalculation 

And your probably right about the 4500T, all the 4000 series familly seems to share the 384Gbps for the stack.

So actually all the 4000 series are blocking (as you said in the inlikely event all the ports are at 100%)

Same for 5530 and below.

But the 5600 series can goes to 1.152Tbps. What is the point of having that capacity ? Saying that a 8 stack will do max 768Gbps (unless I'm still missing someting).

And for the Switch Fabric performance ? What's the definition of that metric ?

[Johan Witters]

There is a big difference between switch throughput (or fabric throughput) or stack throughput..

Switch throughput is the amount of traffic the switch can handle between the physical ports, stack throughput is the mount of traffic it can send up/down over the stacking backplane. This value is the capacity of both the Cascade Up and Cascade Down interface, times 8 (max number of units).

[Paul L]

the big reason why you would want a stack of 5600 to be able to switch up to 1.14tbs, even though that is well beyond what the switch can do is because that stack might be pumping more traffic than what might be physically connected to them. 

example: you might be using a stack of two ERS5632's as an aggregation point for 24, 8 switch stacks.   24*8*48=9216 possible users and servers pumping traffic. And you will want to use 10Ge to connected back to a core 8800 or vsp9000. 

its in that case where you want the aggregate switching power of the switch.

[TankII]

Note, the powerof the 5600 is enough it supports lossless Ethernet, which is what you want with a major iSCSI implementation.

TankII