Hi,

I noticed I'm seeing some Input errors on a gigabit ethernet interface:

70 input errors, 0 CRC, 0 frame, 70 overrun, 0 ignored

the number of input errors seems to increment along with the overrun counter which I assume means that the actual errors are overrun errors.

Does anyone have any tips on finding out what is causing it to overrun?

My first inclination is to assume it is not a huge problem because of the amount of packets that are flowing through this interface:

2367831951 packets input, 247924231216 bytes, 0 no buffer 70 out of 2367831951 is a fairly small number but I wanted to check and see if you all had any thoughts.

thanks,
-Drew

Drew,

Overruns are usually caused by the receiving hardware buffer being "flooded" for lack of a better term because the input rate exceeded the
receiver's ability to handle the traffic.

Darin

Thanks for responding,

As far as you're aware is there a way to check the hardware buffer to see if this is the case, and is this buffer usually per line card, or per slot (both/either?)

-Drew

ASA firewall ?

--
Best regards,
Adrian Minta

Nah this particular instance it is one interface in a 3GE-GBIC-SC in a GSR.

thanks,
-Drew

Unfortunately I don't know of anyway to check the hardware buffer(s), and my "guess" is per line card.

I would also run a "show process cpu" while the overruns incrementing (if you can) to see if the utilization is above 90%. I've heard of this causing overruns in the past. I haven't experienced it myself though.

The only time the CPU utilization is above 10% on this system is when BGP Scanner runs, and it was my understanding that BGP scanner shouldn't cause any issues with traffic.

-Drew

What's the utilization on the other 2 interfaces? I am not familiar with
this specific platform, but it might also be caused by slot/backplane
limitations causing packets to be dropped if the total BW exceeds a certain
(non line-rate) value. I have seen this behaviour on some platforms.

Regards,

Nils Kolstein

Hi,

"Hardware too slow error" - packets arrive in short bursts at line rate,
and your router cannot handle that.

For example, an NPE-G1 will handle packets at, say, 300 mbit/sec if they
come in evenly spaced - packet<pause>packet<pause>packet<pause> - but if
1000 packets arrive back-to-back and then a longer pause, it will overrun
the buffers.

There's not much you can do, except "get a hardware forwarding box"
or "just accept it, and only worry if the errors increase more frequently".

We do some of both :-)

gert
--
USENET is *not* the non-clickable part of WWW!
                                                           //www.muc.de/~gert/
Gert Doering - Munich, Germany                             gert*******
fax: +49-89-35655025                        gert*******

Hi,

Hopefully I'm not completing high-jacking here, but I have seen similar issues on the 4500 w/WS-X4548-GB-RJ45 line cards.  The fabric has 6gbps per slot, so the oversubscription is 8:1.  The best tell tale sign that I'm hitting oversubscription are input errors with no CRC or overruns, like below:

  30 second input rate 6394000 bits/sec, 719 packets/sec
  30 second output rate 722000 bits/sec, 481 packets/sec
     770898484 packets input, 957181248327 bytes, 0 no buffer
     Received 594832 broadcasts (560167 multicast)
     0 runts, 0 giants, 0 throttles
     282191 input errors, 0 CRC, 0 frame, 0 overrun, 0 ignored
     0 input packets with dribble condition detected
     455543646 packets output, 153140605424 bytes, 0 underruns

Is there a more systematic approach to detecting this?  I've gone through some docs and most useful information is geared toward the 6500, such as  http://www.cisco.com/en/US/products/hw/switches/ps700/produc[..]   Currently I have to use a combination of interface statistics and historical Cacti graphs to narrow down over-utilized port ranges.

Thanks,

-ryan

Ryan,
I have similar problems with 4500s so I keep a close eye on the detailed counters.  In particular I watch the transmit drops and also the receive buffer stats.  Pauses frames also indicate a problem in our environment and I would expect in some other environments.  It's a long output but I have always found it very helpful since the reason for the input/output errors are not always evident in a show interface output.


show int counters detail

Port       Tx-Drops-Queue-1  Tx-Drops-Queue-2 Tx-Drops-Queue-3  Tx-Drops-Queue-4
Gi5/34                    0                 0                0                 0
Gi5/35                    0                 0                0                 0
Gi5/36                    0                 0                0                 0
Gi5/37                    0                 0                0                 0
Gi5/38                    0                 0                0                 0
Gi5/39                    0                 0                0                 0
Gi5/40                    0                 0                0                 0
Gi5/41                    0                 0                0                 0
Gi5/42                    0                 0                0                 0
Gi5/43                    0                 0                0                 0
Gi5/44                    0                 0                0                 0
Gi5/45                    0                 0                0                 0
Gi5/46                    0                 0                0                 0
Gi5/47                    0                 0                0                 0
Gi5/48                    0                 0                0                 0
Gi7/1           21257797383                 0                0                 0

show int counters detail
Port         Rx-No-Pkt-Buff     RxPauseFrames    TxPauseFrames   PauseFramesDrop
Gi4/26                    0                 0                0                 0
Gi4/27                    0                 0                0                 0
Gi4/28                    0                 0                0                 0
Gi4/29                    0                 0                0                 0
Gi4/30                    0                 0                0                 0
Gi4/31                    0                 0                0                 0
Gi4/32                    0            107830                0                 0
Gi4/33                    0                 0           346468                 0
Gi4/34                    0                 0                0                 0
Gi4/35                    0                 0                0                 0
Gi4/36                    0                 0                0                 0
Gi4/37                    0                 0             9056                 0
Gi4/38                    0                 0                0                 0
Gi4/39                    0                 0                0                 0
Gi4/40                    0                 0           240746                 0
Gi4/41                 1548                 0                0                 0
Gi4/42                    0                 0          1390048                 0

Nick

Nick,

Thanks, this is what I was looking for.

-ryan

The card in total when I last added everything all up is doing about 1.9Gbps and 1.4Mpps

-Drew

Hi folks,

Does anybody know what causes the router to drop packets as
overrun and what as an input queue drops. There are two show interface
examples of NPE-G1, both with input hold-queue set to 4096. The first
one only shows 153 overrun packets, in the second interface output
you can see overruns together with input queue drops:

GigabitEthernet0/1 is up, line protocol is up
  Input queue: 0/4096/0/58537 (size/max/drops/flushes); Total output drops: 0
  Queueing strategy: fifo
  Output queue: 0/4096 (size/max)
  1 minute input rate 43040000 bits/sec, 6944 packets/sec
  1 minute output rate 23483000 bits/sec, 7180 packets/sec
     2609205324 packets input, 3131277093 bytes, 6 no buffer
     Received 2871721 broadcasts (0 IP multicasts)
     0 runts, 0 giants, 2 throttles
     153 input errors, 0 CRC, 0 frame, 153 overrun, 0 ignored
     0 watchdog, 2871721 multicast, 0 pause input

GigabitEthernet0/3 is up, line protocol is up
  Input queue: 0/4096/4258004/961350 (size/max/drops/flushes); Total output drops: 44638280
  Queueing strategy: Class-based queueing
  Output queue: 6/4096/0 (size/max total/drops)
  1 minute input rate 15685000 bits/sec, 5120 packets/sec
  1 minute output rate 28836000 bits/sec, 5171 packets/sec
     2503236491 packets input, 208082741 bytes, 589462 no buffer
     Received 1329388071 broadcasts (13 IP multicasts)
     0 runts, 12 giants, 960 throttles
     128042 input errors, 12 CRC, 0 frame, 128018 overrun, 0 ignored
     0 watchdog, 1424143105 multicast, 0 pause input

Thanks
Ivan

Do a show controller Gi0/1 | i rx_resource

Chances are the input error count is the same as the rx_resource_error count

This is a microburst issue, and sadly, I know of no way to get around it,
the only solution is buying a router that is able to handle wirespeed Gig

BR,
Sibbi