Activate New NIC in rp8420 HPUX11v1

[IT IS UNIX]

Hello,
At the moment we have only one NIC active in our rp8420. Due to a hughe network load we are planing to activate a second (already installed) NIC. What is the best way to do this to avaoid any problems (routing etc.).

Thanks a lot in advance

Ralf

[Dave Hutton]

Are you planning to use HP APA (HP auto-port aggregation?

http://www.hp.com/products1/serverconnectivity/adapters/apa_overview.html

[Dave Hutton]

Gah, didn't mean to submit. Using APA allows you to load balance and failover, over two cards.

Otherwise my understanding is you can't have a second card on the same subnet. So you would have to have a secondary address to point some of the new traffic to it.

Dave

[IT IS UNIX]

We have a project running to implement this. But this needs some time for testing.
I need at the moment just a second NIC with a dedicated IP for a dedicated connection.

[rick jones]

Generally speaking, unless you are going to aggreagte the two NICs together (assuming they are the same type of network - both 100BT, both GbE etc), it is best to configure each physical (as seen by the transport) NIC into separate IP subnets. Otherwise, the behaviour may not be as you expect/deisre (modulo the use of things like ip_strong_es_model).

In broad terms, assuming your second NIC appears in the output of lanscan, all you need to do is assign it an IP address. You can do that manually with ifconfig, but would have to repeat it on each reboot. The way to deal with that is to edit the file /etc/rc.config.d/netconf - you can either do that by "hand" with vi or emacs or whatnot, or you can use sam.

As for hostnames - you can either add a second IP address to the same hostname in your DNS setup, or (preferably) have three hostnames - one for the system as a whole with both IPs associated, and one specific to each IP, or simply one specific to each IP.

Keep in mind that unless you set ip_strong_es_model, you will only ever have on effective default route no matter how many route add default commands you issue (or entries added to the netconf file).

Now then, what is the definition of "huge" network load here? The rp8420 (iirc) comes with a "core" GbE NIC - are you sure you are close to saturating that already?

[IT IS UNIX]

Hello,

I use glance plus to monitor these systems. When we start the application we get in the Alarm History the red warning for the network. We go up to 5000 packets per second. Is there an other way to see what the throuput of the NIC is?

Thanks all

[rick jones]

Much as I like to see HP book additional revenue from selling you a second NIC, I have to say that Glance can be a bit, well, eager, in reporting network bottlenecks. 5000 packets per second for a GbE NIC may not be saturation.

You need to look at the bytes per second, and the average queue depth of the transmit queue for the NIC before you should believe that the NIC is saturated. If the NIC is _really_ saturated there will be packet drops reported in the output of lanadmin -g mibstats - drops other than from FCS errors or late collisions or whatnot depending on the NIC type.

[IT IS UNIX]

Hi,

we already have the second NIC in the box ;o). I checke the throughput of the NIC. We have 3.5 MB out and 0.3 MB in rate. That is not so much but we already got a yellow warning. It seems that the settings for Glance are a little strange. We get yellow warning between 50% and 90% utilisation. But what are 90%? are these the values from 100 MB NIC or will this be calculated for each card type?

[rick jones]

If the Glance version is old enough, they could be for a 10 Mbit/s NIC :) It is also possible they could be for 100BT. Somewhere in a config file you should be able to find the settings. If it isn't mentioned in the glance documentation, you could tusc the start-up of glance (I'd suggest the character version rather than the X11 gpm if tuscing) and see which file it opens and reads. Heck, it might even keep the file open, in which case, looking at the open files for glance, in glance may show it :) Or lsof could be used.

[Steven E. Protter]

The cardinal rule of HP-UX is:
Without APA(mentioned above) you can't have two NIC cards up and active on the same subnet.

I've tried it. The results are catastrophic.

So don't try that in production.

If you don't want to go APA, upgrading to Gigabit will help some.

:-)

SEP

[IT IS UNIX]

Ok, I found this in /var/opt/perf/alarmdef:

# The Network bottleneck symptom default relies on general throughput
# metrics. Not all network interfaces report collision data. To be
# useful as a bottleneck indicator, the rate thresholds should be
# adjusted based on values seen in historical data for a particular
# system or network. For example, 10mbit networks cannot handle as
# high packet rates without a bottleneck than can 100mbit networks.
symptom Network_Bottleneck type=NETWORK
rule GBL_NFS_CALL_RATE > 500 prob 25
rule GBL_NET_COLLISION_PCT > 10 prob 10
rule GBL_NET_COLLISION_PCT > 25 prob 20
rule GBL_NET_COLLISION_PCT > 50 prob 30
rule GBL_NET_PACKET_RATE > 500 prob 10
rule GBL_NET_PACKET_RATE > 1000 prob 15
rule GBL_NET_PACKET_RATE > 3000 prob 20
rule GBL_NET_PACKET_RATE > 5000 prob 25
rule GBL_NET_PACKET_RATE > 9000 prob 25

My question is now: What must be change to have the right values for 1 Gbit NIC? What does the "prob 10" or "prob 20" (example ) means?

Thanks a lot

Ralf

[rick jones]

the "prob N" stuff says to add that much to the probability of there being a network bottlenect. as the probability goes higher, glance goes from green to yellow to red and such.