Disclaimer
The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.
Liability Disclaimer
In no event shall Author be liable for any damages whatsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.
Posting
JosephDoherty, I take it that the "bridge" in your road example (which I like very much by the way) is the switches backplane. This example makes sense to me. So, my question is when Cisco refers to the switching bandwidth, are THEY referring to the backplane of the switch? I see for a 3750-X that the switching bandwidth is 160 Gbps. That is on a 24 port switch, which seems well above the, what I think would be 96G needed for full capacity for all ports? Which I would deem as a good thing. So, do I understand "switching bandwidth", as you guys see it?
On the 3750-X datasheet, we see:
Table 9. Cisco Catalyst 3750-X and 3560-X Performance Specifications
Performance Numbers for All Switch Models |
Switching Fabric | 160 Gbps |
This, as described, is the bandwidth capacity all these switch series internal switching fabric, i.e. the bandwidth capacity between switch's ports. BTW, backplane might refer to the switch's internal bandwidth capacity, but it has other meanings too. I would avoid using it in this context.
For 24 (duplex) gig ports you need 48 Gbps. However, these switches often support an optional module that can support two 10g (duplex) ports, so then your bandwidth requirement is 88 Gbps ((24 + 20 [dual 10g]) * 2 [duplex]). (For the 48 gig ports, you would need 136 Gbps.)
In Andras's post, he mentions bandwidth of the ring ports (note: this would be only for the 3750-X). I'm unsure Cisco "counts" them against the fabric capacity, as Andras supposes, but assuming they do, each port is 16 (duplex) gig. So a 24 port 3750-X would need (24 + 20 + 32 [dual 16g]) * 2 = 152 Gbps, while the 48 port 3750-X would need (48 + 20 + 32) * 2 = 200 Gbps.
The described 160 Gbps is more than enough for any 3560-X or any 3750-X not stacked. However, if we do need to count StackWisePlus port bandwidths against the fabric, the 48 port 3750-X has insufficient bandwidth to support full concurrent port utilization.
Again, Cisco isn't clear about whether StackWisePlus capacity should be counted against the provided fabric bandwidth, but if we look at the original 3750 series, its switching fabric bandwidth is described as 32 Gbps. Considering StackWise bandwidth is also described as 32 Gbps, I suspect this bandwidth was not intended to "count" again the switch's fabric bandwidth. As the original StackWise is described as placing all switch traffic on the ring, I suspect this bandwidth is in "parallel" to the fabric's bandwidth.
When the total fabric capacity exceeds port capacity, there's no benefit to performance as it will remain unused.
As to why all the 3560-X/3750-X series have the same fabric bandwidth, likely it's a one-size for all design. Different models probably just "connect" to the same basic fabric chip set. (Somewhat similar to old 6500 SFMs [switch fabric modules]. I.e. they provided a fabric that supported up to 256 Gbps.)
Regarding "forwarding rate", that's a measure of how many packets per second the switch can process for certain sized packets. When packet's size isn't described, today it's usually denoted for minimum size Ethernet packets, i.e. each 64 bytes.
Minimum size Ethernet, to run at gig rate, is 1.488 Mpps. Unlike fabric, you don't need to account for duplex as one port's in is another port's out.
So for 24 gig ports, and for the optional dual 10g ports, we need (24 + 20) * 1.488 = 65.472 Mpps to support full rate.
The spec sheet notes the 24 ports switch's Mpps as 65.5, which would be a rounded number of the 65.472 Mpps just calculated.
Notice how this would be fine for all the ports excluding the StackWisePlus ports. (Again, why I believe they are not "counted" as part of these bandwidth/Mpps specs.)
