Too many Inbound discards

[Sajith V Mannadiar]

Hello there,

While analysing the tcpdump output for an intermittent application error, we found a few instances of packet losses.

When we checked using lanadmin, we found lots of 'Inbound discards' being recorded on the problem server.

We have 4 identical application servers running Oracle 9iAS on HPUX 11i, but only one of them records around 300,000 Inbound discards and we also found that whenever the user session is on this particular server, the session hangs..

All the servers are identical (in terms of hardware, configuration, application s/w version etc)

Any thoughts?

Please help

Thanks,
Sajith

[Ashwani Kashyap]

Is your network card configured in half duplex mode , then configure it for full duplex .

It also helps of the seetings on your netork card matches the settings on the switch port where it is connected to . Also if its not a gig-e card then turn off auto negotiation at both ends .

[HGN]

Hi

Try and check the lan interface setting speed
# lanadmin -x 0 (for lan0)
Current Config = 100 Full-Duplex MANUAL

Make sure the lan card and the switch are at the same speed setting.

Still if the problem persists then you have to start trouble shooting by moving this to another interface (if you have any free cards) or change the lan cable or change the switch port.

Good Luck

Rgds

HGN

[Blofeld]

Sajith
On top of what the other guys said you may also want to have the Network team look into port statistics of where your server is plugged into. I have run into instances where Cisco ports go bad before they register bad on the card

[Sajith V Mannadiar]

Thank you Folks,

The ethernet card is configured at 100 Mbps Full Duplex

I shall check the switch ports... Is there a way to test that from the Unix server?

Thanks,
Sajith

[Sajith V Mannadiar]

Can someone tell me why the 'Inbound errors' are 0 and the 'inbound discards' is high?

If there are errors in the switch port or cable, shouldn't it have given 'inbound errors'?

Regards,
Sajith

[Jim Keeble · Accepted solution]

The inbound discards statistic is an overflow condition on the card. Packets are coming in from the network faster than they can be tranferred to the IP layer above.

Generally, the problem is on the back end. Packets are not coming too fast, something is preventing the card from being able to transfer the data up to the stack.

Look for lots of activity on a bus that is shared by the network card, cpu utilization, memory pressure, that sort of thing.

There could also be a HW problem either on the I/O bus or on the cards interface to it.