Network Bottleneck

[Jackie Wynn]

I'm getting network bottleneck alerts of 100% every few minutes lasting 5 minutes each time. My alarmdef file is set for a 100mps system since the customer's network is not a gigabit system. Does anyone have any suggestions?

[harry d brown jr]

Jackie,

What processes are running during that period? Do you have lsof?

http://hpux.cs.utah.edu/hppd/hpux/Sysadmin/lsof-4.61/

Also, check crontab to see if any job is firing up every five minutes.

live free or die
harry

[Ceesjan van Hattum]

What does 'glance' tell you at those moments?
press l 'network by interface
press N 'NFS GLobal Activity
press A 'Alarm History

-Ceesjan

[Rusty Sapper]

Hi,
You should also check the syslog during one of the bottlenecks to see if anyone is connecting remotely to the server. The may be pulling or pushing large files during that time.



HTH

Rusty

[Shannon Petry · Accepted solution]

Depends on the topology. I'll assume from what you wrote that it's a single 100Mbps NIC? I'll assume you have checked for network errors? Collision rates? etc...
Is it in a switch or hub?
Do you have multiple NIC's?

What applications are using the network?
What else is on the network with the server?

Basically, if the customer does not want to invest in GIG or ATM, then you will need to plan out (very well) multiple NIC's and subnets. Not an easy task!
I.E.
Oracle database, FTP Server, and local access.
Nic1. 10.10.10.1
Nic2. 20.20.20.1
Nic3. 30.30.30.1

This would segment your network requiring only 2 new switches (and not large ones), and re-route your traffic.

I'd recommend however a gig card. Several switch vendors have a single gig port for servers, and the rest of the clients at 100mbps so wont kill the budget. This will leave room for growth as well.

Regards,
Shannon

[Sandip Ghosh]

Can you check the speed settings in your NIC and Switch? If there is a difference then it may happen. Set both the port as 10mbps or 100 mbps, according to your configuration and see the performance.

Sandip

[Jackie Wynn]

Shannon,
Nothing has changed in the past few weeks with this customer. Everything was fine up until last Monday. They're only running 100mbs cards. I researched their network usuage via mrtg. The most that they are bursting up to is about 80mb.

[rick jones]

it is my expereince that even today, glance is a bit, well, paranoid about network packet rates.

while glance is alarming on network, use lanadmin to check the queue length for tne NIC (or NICs) and see if it is constently non-zero. also, check for outbound discards or inbound discards, those could indicate that indeed there is a network bottleneck.

[Doug Grumann]

Sometimes I need to remind Rick that he had the glance team put in his favorite network bottleneck metrics. In recent versions, the following should all be available via HP-UX gpm's Choose Metrics screens:

Global Info: Average queue length, collision deferred and error packets rates and percentages, IP fragments, forwarded datagrams, and reassembly required counts.

Network detail by Interface: Packet and byte rates, collisions and errors, and queue length.