[c-nsp] ISR G2 Interface RX Performance

[Richard Clayton]

On 25 January 2013 23:11, Nathanael Law <Nathanael.Law at aimco.alberta.ca>wrote:

> Hello all,
>
> We're having some issues with a 3925 and real-time UDP traffic bursts.
>  The bursts
> are approximately 1500 packets long and are sent in 5.7 ms for an
> effective rate
> of ~250 kpps (~375 Mbps).  The steady state traffic on this connection is
> < 10kpps.
>
> Physical Topology
> =================
>
> +------+                      +------+
> |      |                      |      |
> | 3750 | gi2/0/2 ------ gi0/0 | 3925 |
> |      |                      |      |
> +------+                      +------+
>
> Packet captures have shown that the 3750 gi2/0/2 interface has no issues
> sending
> the entire burst; however, packet captures on both the receiving host and
> the
> 3925 shows that only about 1/3rd of the packets show up on the gi0/0
> interface.
> The overrun counter increases slightly with each burst.
>
> 3925#show interfaces gi0/0
> GigabitEthernet0/0 is up, line protocol is up
>   Hardware is PQ3_TSEC, address is 70ca.9bb5.7a80 (bia 70ca.9bb5.7a80)
>   Description: Uplink to core switch 2
>   MTU 1500 bytes, BW 1000000 Kbit/sec, DLY 10 usec,
>      reliability 255/255, txload 12/255, rxload 1/255
>   Encapsulation 802.1Q Virtual LAN, Vlan ID  1., loopback not set
>   Keepalive set (10 sec)
>   Full Duplex, 1Gbps, media type is RJ45
>   output flow-control is unsupported, input flow-control is unsupported
>   ARP type: ARPA, ARP Timeout 04:00:00
>   Last input 00:00:00, output 00:00:00, output hang never
>   Last clearing of "show interface" counters never
>   Input queue: 0/75/176/0 (size/max/drops/flushes); Total output drops: 0
>   Queueing strategy: fifo
>   Output queue: 0/40 (size/max)
>   5 minute input rate 5331000 bits/sec, 2511 packets/sec
>   5 minute output rate 48693000 bits/sec, 5037 packets/sec
>      40386885831 packets input, 7413412230506 bytes, 0 no buffer
>      Received 3800689 broadcasts (0 IP multicasts)
>      0 runts, 0 giants, 111 throttles
>      82202 input errors, 0 CRC, 0 frame, 82202 overrun, 0 ignored
>      0 watchdog, 3800681 multicast, 0 pause input
>      80802723127 packets output, 110122764930458 bytes, 0 underruns
>      0 output errors, 0 collisions, 1 interface resets
>      0 unknown protocol drops
>      0 babbles, 0 late collision, 0 deferred
>      0 lost carrier, 0 no carrier, 0 pause output
>      0 output buffer failures, 0 output buffers swapped out
>
> The only other statistic of note seems to be the rx_overflow_err on the
> gi0/0
> interface:
>
> Internal Driver Information:
>  throttled=111, enabled=111, disabled=0
>  rx_coalesce_failed=0, rx_framing_err=0, rx_overflow_err=1726086,
> rx_buffer_err=65
>  rx_no_enp=0, rx_discard=0
>  tx_one_col_err=0, tx_more_col_err=0, tx_no_enp=0, tx_deferred_err=0
>  tx_underrun_err=0, tx_late_collision_err=0, tx_loss_carrier_err=0
>  tx_exc_collision_err=0, tx_buff_err=0, fatal_tx_err=0
>
> From this it seems that the actual gigabit physical interfaces on the 3925
> cannot
> handle even 11% of line rate (2 Mpps for a 1 Gbps connection @ 64 byte
> packets).
> I knew that the processor can't handle that, but I expected the interface
> itself
> to come at least a little closer given that CEF on the 3925 can supposedly
> handle
> 833 kpps without any features turned on.
>
> Does my analysis seem accurate?  If not, any pointers in the right
> direction would
> be appreciated.  If so, what Cisco routing hardware would be minimally
> required to
> support line-rate 1 Gbps input (the 3925 is a WAN router that basically
> passes
> traffic off to one of our MPLS providers)?  Would the interfaces on an
> ASK1k (ESP5)
>  do the job?
>
> We do have a TAC case open for this issue, but they have been unable to
> provide
> documentation on the limitations of the physical interface so far.
>
> Documents that seem related:
>  - https://supportforums.cisco.com/docs/DOC-2613 (doesn't reference
> anything as new
>    as the ISR G2s, but I figured it may still apply)
>
> Thank you,
>
> Nathanael Law
>
> _______________________________________________
> 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/
>


Nathanael

I have some lab stress test results for the whole ISR G2 platform which I
can share with you if you like, a quick look at the 3925 with a traffic
profile of G711 and packet marking (you would probably have marking in a
voice environment with QoS applied) shows the cpu at 75% whilst it is
passing 280Mb.  The bursts could be topping the cpu out although my
realtime tests were with voice codecs and your realtime traffic is
1500bytes so the cpu should be at a lower rate for the same amount of
bandwidth.  Graphing the cpu isnt going to help as its such short bursts
but the overuns are a real indicator that the cpu is stressed.
Shaping would normally be the answer to control bursty traffic but shaping
realtime will cause issues with the stream.  If you want to allow this
traffic then you need a cpe with a bit more grunt, the 3945 is the next
step up and is about £1000 more but you need to make sure this would defo
fulfil your needs.

What is the realtime traffic in question?

Thanks
Sledge