forwarding performance rate

[bluesea2010]

Hi,

I have 28  access (stack with 4 ) is connected with 10 g Uplink  to one core switch  ?

what switching capacity should i look for , and forwarding rate  500 Mpps is enough ?

Thanks 

[Joseph W. Doherty]

Minimum size Ethernet needs 1.488 Mpps for 1 Gbps.

So, take your maximum expected concurrent bandwidth forwarding need (don't forget duplex), and you can figure out your PPS need.

BTW, maximum (standard) size Ethernet only needs 81.274 Kpps per 1 Gbps.  We usually use minimum for calculation as its the worst case.

[bluesea2010]

Hi,

Thanks a million , now to reach   switching capacity for the same  core switch , how can I do that ?

Thanks

[Joseph W. Doherty]

Basically, the same applies to all network devices, regarding PPS needs.

BTW, most (this century) Enterprise (or better [e.g carrier grade]) switches' PPS can often handle all ports, concurrently, running at wire-speed even for all minimum sized packets.  However, one "gotcha" with switch performance, is careful analysis of other features, like sufficient buffers to handle queuing and/or other features you might need, like NAT, QoS, tunnels, etc.

[bluesea2010]

Hi @Joseph W. Doherty 

Thanks for the reply 

the below taken from the 9500 datasheet , 

1) Up to 36 MB of unified buffer per ASIC
2 ) 80MB of dedicated low-latency buffer, with up to 8GB of HBM buffer

Could you please brief  how both impact the switch performance 

And second thing  How do we size " switching capacity " ( not forwarding performance ) 

Thanks 

[Joseph W. Doherty]

"Could you please brief how both impact the switch performance."

If insufficient, you'll often see an increased drop rate during bursts of congestion on interfaces.

"And second thing How do we size " switching capacity " ( not forwarding performance ) "

Ah, that's really a "black art", as for starters you would really, really (really) need to know your traffic requirements (and most, if not almost all, times we don't).

So, what's often done, we use less expensive/capable L3 switches closer to the edge, and more expensive/capable L3 switches closer to the core (or the core); then we cross our fingers.

In the real world, we're often asked can't we use a less expensive L3 switch "here" (or "there").  (I.e. either less expensive L3 switch models, from the same vendor, or less expensive L3 switches from another vendor.)  If not, justify why not.  Truly, a) it's hard to (again) really, really (really) know your needs, and b) we often over estimate equipment needs because we don't really know what we need.  So, we generally "over provision", i.e. select equipment to avoid coming up short in actual operations.

Unfortunately, I don't know of a nice easy "cookbook" approach to get "just what's needed".

Perhaps, someone else will reply on this subject to with a better answer.

[bluesea2010]

Hi @Joseph W. Doherty 

Can you pl

cisco 9500 48x10/25g +4x100g

10/25 =37.2 (25*1.488)
48*37.2 =892.8

for 100gig =148.8
4x100 =1041.6
total forwarding performance require =1934.4 Mpps
int the datasheet it shows  upto1bpps . The actual require is 1934.4 why cisco says upto 1bpps ?

one non-cisco vendor with the same spec shows lower than 1934.4 Mpps

Thanks

[Joseph W. Doherty]

"int the datasheet it shows upto1bpps . The actual require is 1934.4 why cisco says upto 1bpps ?"

It might just simply mean the device cannot provide the almost 2 Bpps you calculated.

However, remember you're computing PPS for all ports running concurrently as 100% utilization and also using minimum packet sizes, i.e. the very, very worst case.  Real world, for most usages, you'll never come close to that resource demand need.

Years ago, many switches also could not provide for very worst case usage load either, and it was very seldom a problem.  The later platforms that could, were more about bragging your switch could than meeting an actual real-world need.

[bluesea2010]

Hi @Joseph W. Doherty 

The actual  for 9500 48 port  switch 1934.4 Mpps( not bpps )

in datasheet is upto 1 bpps 

Thanks  

[Joseph W. Doherty]

Sorry, now confused by your last posting.

Ah, wouldn't 1,934.4 Mpps be the same as 1.9344 Bpps?

Didn't you compute PPS needs as 1934.4 Mpps or 1.9344 Bpps?

If datasheet has up to 1 Bpps, again, perhaps that's all it can deliver, again, real-world, plenty for normal usage.

[bluesea2010]

Hi

Sorry for the confusion , as per the calculation based on the port  (48X10/25 + 4 x40 /100)  it supposed to have 2 bpps

but in the datasheet it says only  1 bpps

That was my question 

[Joseph W. Doherty]

Okay, well again, over the history of switches, they often has less capacity than what would support the most demanding PPS rate.

Just as an example, one of the Catalyst 3750G switches could be had with 48 copper gig ports and 4 (gig) SFP ports, i.e. total of 52 Gbps, which would need a PPS rate of (about) 77 Mbps and a fabric supporting 104 Gbps.  However, the specs for that switch were a maximum of 38.7 Mpps and a 32 Gbps fabric.  I.e. this model switch could not support all packets sizes, at full wire-rate, on all ports, concurrently!

Conversely, though, the (about) same time Catalyst 4948, which had 48 (active) gig ports, provided 72 Mpps and 96 Gbps fabric, both exactly what was needed for wire-rate, on all ports, concurrently, regardless of packet size.

Why the two switches, one that did not have the wire-rate for all ports, concurrently, and one that did?  Well, the 4948 was much, much more expensive than the 3750G.

I'm sure Cisco could design a 2 Gpps switch, but it too would likely be much, much more expensive than the 9500.  So, do you wish to pay more, for capacity you're unlikely ever to need?  I presume, Cisco believes most customers would not.