[c-nsp] GigE woes

[Aaron]

I wouldn't expect DWDM/Trasmission gear to have that kind of impact. That
equipment doesn't have that kind of intelligence. It has to be the end
equipment.

On Sun, May 16, 2010 at 03:50, Alexander <ecralar at hotmail.com> wrote:

> Tim,
> I am willing to bet that DWDM/transmission eqmt in between is waiting to
> "lump" several small packets together before attempting to send.
> Hence ping timeouts with small packets and no timeouts with large ones.
> Seen this before with ATM cells in Juniper ATM-1 PIC.
> HTH
> Regards
> Alex
>
>
> --------------------------------------------------
> From: "Tim Durack" <tdurack at gmail.com>
> Sent: Friday, May 14, 2010 7:53 PM
> To: <cisco-nsp at puck.nether.net>
> Subject: [c-nsp] GigE woes
>
>
>  I've got a crazy GigE circuit that's having problems:
>>
>> RTR-1#ping vrf v10 10.241.1.10 repeat 10000 df-bit size 9000 timeout 1
>>
>> Type escape sequence to abort.
>> Sending 10000, 9000-byte ICMP Echos to 10.241.1.10, timeout is 1 seconds:
>> Packet sent with the DF bit set
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!
>> Success rate is 100 percent (587/587), round-trip min/avg/max = 8/18/408
>> ms
>>
>> RTR-1#ping vrf v10 10.241.1.10 repeat 10000 df-bit size 200 timeout 1
>>
>> Type escape sequence to abort.
>> Sending 10000, 200-byte ICMP Echos to 10.241.1.10, timeout is 1 seconds:
>> Packet sent with the DF bit set
>> ..!..!..!.!....!..!..!.!!..!...!..!..!..!..!.!!...!.!..!..!....!..!..!
>> ..!..!....!..!..!..!.!.!..!....!.!..!..!..!..!....!..!..!..!..!.!.!..!
>> .!..!..!..!..!....!..!..!..!..!....!!..!.!..!....!.!!....!..!..!..!..!
>> ....!.!..
>> Success rate is 32 percent (72/219), round-trip min/avg/max = 4/160/1180
>> ms
>>
>> RTR-1#ping vrf v10 10.241.1.10 repeat 10000 df-bit size 9000 timeout 1
>>
>> Type escape sequence to abort.
>> Sending 10000, 9000-byte ICMP Echos to 10.241.1.10, timeout is 1 seconds:
>> Packet sent with the DF bit set
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
>> Success rate is 100 percent (255/255), round-trip min/avg/max = 12/16/228
>> ms
>>
>> Ping with large frame size appears to pass. Small frame size results
>> in packet loss and unusually high rtt.
>>
>> Additionally, traffic must be present in both directions for any
>> traffic to cross the link. I have to start the ping on both routers
>> simultaneously. If the ping is stopped on one side, the ping from the
>> other side stops returning.
>>
>> Have mostly ruled out my equipment, but the carrier doesn't believe
>> there is anything wrong on their side. They strap a GigE tester across
>> the path and of course it works just fine. But try convincing arp/ospf
>> to work across such a link.
>>
>> The link appears to be a muxponder over the carriers DWDM network, so
>> it is mostly optical, possibly only two OEO points at the customer
>> hand off.
>>
>> Anyone seen anything like this before?
>>
>> --
>> Tim:>
>> _______________________________________________
>> 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/
>>
>>  _______________________________________________
> 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/
>