C7206VXR NPE-G1 & G2 Input Errors & Overruns

routeinet · ‎08-15-2010

Hello All, I'm hoping that there are other enterprises out there experiencing same problem as we're having with "Input Errors" and "Overrun" on

C7206VXR NPE-G1s and G2s. All of our C7200VXRs are experiencing "Input Error" & "Overrun" constanly incrementing on GigE interface.

It seems there's no relationship to amount of traffic traversing on the GigE interface.

But for the purpose of this discussion we're averaging 44mbps and max burst at 86mbps for last year or so on Trunked GigE interface on C7206VXR NPE-G1 & G2 connecting to 6509E(Multiple pairs of NPE-G1s and Pair of NPE-G2 in the network). TAC says that we're oversubscribing the hardware limitations but I'm not convinced that we're over subscribing the hardware limitations.

What is the hardware limitations on GigE interface on C7206VXR NPE-G1 & G2 ?

Giuseppe Larosa · ‎08-15-2010

Hello routeinet,

being connected to more powerful C6509 these interfaces are exposed to possible buffer overruns that are counted as input errors

However, given the low average traffic volume (86 Mbps) what is the ratio between input errors and total received packets?

if it is in the order of 1 packet lost every 10^5 there is no real impact.

if the ratio is lower then you should consider flow control for gigabit ethernet interfaces

What TAC can have said is that if the other device sends N packets in a row at line speed your interface may not be able to handle all the packets if its low level buffers can handle M packets with M < N

So in some cases some packet can be lost and counted as input error overrun on C7200 GE interface.

Now, as I have noted to understand if this may have an impact what counts is the ratio between these errors and totally received packets

Hope to help

Giuseppe

.

routeinet · ‎08-15-2010

Hello Giuslar,

Thank you very much for your comments.

We're seeing 13099 input errors / 300923430 packets input.

That's 1 input errors per 23K pkts.

I understand that connected GigE int on C6500 sends burst of traffic to GigE on C7200.

But I'm having hard time understanding at what rate should I consider that I'm seeing bursting.

Just to clarify my understanding,

Are you saying that by nature of these hardware differences, C6500 sends data faster than C7200 can handle which causes input errors ?

Or Is it "bursting" nature of traffic causes the input drops,

The Input error incrementation is almost non-discriminating to amount of traffic it's receiving.

We're just seeing constant "Input Error" incrementation. (reliability 255/255, txload 5/255, rxload 4/255)

Do you know If there's a limit to bursting limitations on C7200, what could that be ?

Can C7200 platform handle 1k pkts/sec without any error or 10k pkts/sec ?

If C7200 can house OC3 or DS3 card, Shoud It not be able to handle minimum of DS3 or OC3 rate ?

or Am I talking about completely different technology ?

On flow control note,

Is this common practice to turn the flow control on C6509E connecting to C7206VXR GigE interface ?

(Flow control status shows off - input / on - output)

Giuseppe Larosa · ‎08-16-2010

Hello Routeinet,

the good news is that the error rate is low

>> That's 1 input errors per 23K pkts.

the problem is caused by burstiness not by traffic volume I have been able to test NPE-G1 with 500 Mbps of traffic between two GE ports using a traffic generator with no losses, but traffic generator traffic is regular (too much regular in some cases).

You could try to enable the sending of pause frames on C7200 side as a way to stop incrementing these counters, that is to enable flow control out on the C7200 interface

in order for this to work well you need autonegotiation ( no hard coding of speed and duplex)

Or you could stay in this scenario as the practical effects are zero even on UDP flows (TCP flows can retransmit the missed packet because they have session control).

Hope to help

Giuseppe