1000Base-T Adapter and Cisco Core switch---- Packet drop

[Rgomes]

Hi all,

We have L3000 server having a 1000B-T card connected to cisco core switch 6509. Whenever we ping the server IP address( 10.10.20.27) from our PCs, we see massive amount of packet drops. Is it compatibilty issue with two hardwares? Anyone experienced this before?

Pls see the attachment for further information.

Thanks and regards,
Richard

[Michael Schulte zur Sur · Accepted solution]

Hi,

make sure that port and nic settings are the same. speed/duplex, auto sensing/hard setted.

Michael

[Rgomes]

Hi Michael,

Thanks for your reply. These are configured:
# lanadmin -x 1
Speed = 1000 Full-Duplex.
Autonegotiation = On.

Best regards,
Richard

[Michael Schulte zur Sur]

Is the setting on the switch auto sensing?
Can you check for collisions on the switch? Have you tried a different port?

Michael

[Rgomes]

Hi Michael,

Switch port is autosensing and I tried with another port also.

nettl does not show any error message while I am ping'ing.

Thanks,
Richard

[Steven E. Protter]

If the switch is set to auto sense and the card to autonegotiate there are a few things left the check:

1) Cabling
2) Bad swithc port
3) IOS problem on the switch or unsupported configuration
4) Bad network card.

dmesg
use cstm or xstm and test the card thoroughly.

SEP

[Jeff Schussele]

Hi Richard,

Gig-E is a "special" beast. ALL Gig-E connections - whether they be fibre or copper - *must* be auto-negotiate when operating at Gig speeds. Therefore, I would suspect a "problem" somewhere - probably a physical fault - i.e. bad/damaged cable, NIC or port.
I would also verify that virtual settings such as VLAN & such are set up poperly.
And finally if everything "appears" to be set & functioning correctly, that you start tracing the traffic with sniffers or nettl or such.
And make damn sure you're up to date on all PHNE patches.

Rgds,
Jeff

Oh & check the *actual* speed/settings with
lanadmin -x Y
where Y = PPA - i.e. 0 for lan0 or 3 for lan3

[Michael Schulte zur Sur]

Hi Richard,

have you checked the software version on the cisco? Could you reboot the switch?

Michael

[Bill Hassell]

Another possibility: try arp -a and look for a duplicate MAC address. If there are lots of entries, pipe it to awk and sort:

arp -a|awk '{print $4" "$1" " $2}'|sort

If you see a duplicate, this will cause dropped ping replies. This occasionally happens in ServiceGuard environments where the MAC address is changed on LAN cards during bootup. You can use arp -d to delete the bogus entry as a quick test. ping should work much better now. Then fix the problem machine, preferably not connected to the network.

[Rgomes]

Hi All,

Thanks to you all for your great inputs! I am checking and trying to understand this problem. But still no luck. I tried with 2 other cable ( cat 6e), by changing switch port but ..... :(

I understand the port, to which my server's giga card is connected is a member of Vlan, and core switch also playing a role like DHCP server. But again still haven't found any clue.

Can you recommend any trial network tracer s/w?

Thanks to you all again,
Richard

[Steven E. Protter]

Try this, it has the advantage of being free and very good.

http://hpux.connect.org.uk/hppd/hpux/Gtk/Applications/ethereal-0.9.15/

SEP

[Rgomes]

HI SEP,

Thanks for the link, I'll try with that.

Bill, where is your logo?? :)

regards,
Richard

[Rgomes]

Hi All,

I am thinking about ping behavior and reverse lookup zone file.

When I put 'ping -a 10.10.20.27' it returns with another fqdn name( see attachment), which is not associate with HPUX box( host name is mis). This reverse record is old one having 10.10.20.27 with previous entry of 'gp-pc-1320.mydomain.com'.

Is it might be an issue?

regards,
Richard

[Bill Hassell]

ping -a ???

Oh I see, this ping isn't from HP-UX to HP-UX. So the first thing to determine is exactly where your PC is getting it's IP information. You may have an /etc/hosts (actually, /drivers/etc/hosts) file on the PC or perhaps WINS is getting involved (aren't PCs fun?). You may have some DNS issues but these won't drop packets occasionally. What does nslookup 10.10.20.27 say about the source of the hostname? (I'm talking Unix here) Or better yet, use nsquery as in:

nsquery hosts 10.10.20.27
nsquery hosts 10.10.20.27 dns

and also do a traceroute to see if there are strange routes involved:

traceroute 10.10.20.27

[Rgomes]

Hi Bill,

Yes I am issuing ping from client PC.
There is no DNS service configured in this machine or any of the unix machine. We are using /etc/hosts file for name resulation.

I am primarily concerned about the ping behavior from PC to hp-unix box.

Thanks and regards,
Richard

[Michael Schulte zur Sur]

Hi,

have you looked into drivers/etc and what about Bill's wins idea?

Michael

[Rgomes]

Hi All,

Finally we have solved this issue.

Just to recall:
OS ver: HPUX 11i
Gigabit NIC: HP A6825-60101 PCI 1000Base-T Adapter

From HP it was stated that:
" There is a known issue with the A6825A gigabit card. (Service note: A6825A-01)

Gigabit Ethernet interfaces using igelan driver may generate TCP/UDP Checksum Errors thereby causing packet drops."

For this to fix, there are 2 patches.One is 'site specific' and the other is 'general released'. In our case, only general release patch was needed, which is PHNE_29631. Now there are no packet drops.

This is for future reference.
Thanks to you all who shared with me their valuable time.

Richard

[Sriram Narayanaswamy]

Hi Richard,

I'm glad your problem has been solved.
But there's a small correction: pings generate
ICMP packets which do not require the TCP/UDP
checksumming feature. Therefore, while PHNE_29631
may have fixed your problem, your problem is
not the one mentioned in the service note
you've referred to.

regards
Sriram

[Jeff Schussele]

Hi (again) Richard,

You should probably be looking over the switch/router logs.
A *lot* of routers are configured such that under load they will drop ICMP packets because they are *low* priority. This may be what's occurring. So check the router logs in path.

Rgds,
Jeff

[Rgomes]

Hi Sriram and Jeff,

Thanks for your correction.You are right. After 2weeks, I am still having packet drops. Why?? I don't know. Hmmm, let me see all the necessary log files. :(

Pls ignore my previous mail.

Thanks and regards,
Richard