[c-nsp] Maximum traffic on Gigabit Ethernet

[John Gill]

Manaf,
Your rate may be fine depending on packet size, for example 65 byte 
frames are less efficient in the hardware path and buffers than 64 
bytes.  You wouldn't expect more than about 760Mb/s in the best case 
with that size frame.  With no congestion and one continuous input, 
9000B frames can achieve about 990Mb/s.

So, the question here: is this a problem or an observation?  What rate 
do you expect?  Has anything changed?  Let's see if you have drops, also 
if you have CPU-switched flows, that will cause noticeable latency and 
will limit throughput.  Sniffing the traffic may show retransmissions, 
which would mean drops occur somewhere in the path.

The only thing you can do at this point, without changing the traffic 
flows (or can you?) is to look at drops and adjust buffer settings if 
desired.

For dropping traffic, you can look at "show int gi 5/2" and see if Total 
output drops is increasing.

Also look at "show queueing interface gi 5/2" and at the bottom you 
should see packets dropped on transmit and dropped on receive.  See if 
these are incrementing.

Finally, if you are getting software drops somewhere on input, you need 
to look at all other L3 interfaces that could possibly send traffic here 
(including this one) for input queue drops in software:
"sh int | inc is up|queue"

Interesting output would look like this:
...
TenGigabitEthernet2/3 is up, line protocol is up (connected)
  Input queue: 3/2000/189/15 (size/max/drops/flushes); Total output drops:0
...
TenGigabitEthernet2/3 is up, line protocol is up (connected)
  Input queue: 18/2000/264/16 (size/max/drops/flushes); Total output 
drops: 0

Notice how drops and flushes have incremented here, indicating 
congestion to the software path (CPU punt).  There are many reasons 
traffic might be punted, but let's see if that's the case first.

Regards,
John Gill
cisco

On 10/4/11 5:49 AM, Nick Hilliard wrote:
> On 04/10/2011 07:38, Manaf Al Oqlah wrote:
>> Normal Configuration:
>>
>> interface GigabitEthernet5/2
>> ip address x.x.x.x
>> no ip redirects
>> no ip proxy-arp
>> media-type rj45
>>
>> MTU is default 1500
>
> The GE ports on a VS-S720-10G-3C supervisor are each provisioned with 9.6MB
> RX and 8.1MB TX buffers.  This is quite a respectable amount of buffer
> space for a GE port, although if you have mls qos enabled, the default TX
> configuration is 1p3q4t, configured as:
>
> SP: 1.2MB
> Q3: 1.2MB
> Q2: 1.6MB
> Q1: 4.1MB
>
> This may not suit your traffic profile if everything is in a default
> priority class, as you'll end up putting everything into q1.
>
> reference paper:
>
>> http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper09186a0080131086.html
>
> Nick
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>