Traffic between two servers.

[Gulam Mohiuddin]

We would like to find out the network traffic between two HP-UX Server, one is PeopleSoft's backend Oracle database server and the other one is PeopleSoft Application Server.

Currently we are experiencing some network bottlenecks and trying to investigate how to improve network performance between those servers.

Thanks,

Gulam.

[Pete Randall]

Gulam,

When someone mentions network bottlenecks, the first thing to check is that speed and duplex settings are properly configured on both the switch and the hosts. HP-UX does not do well with auto-negotiation and needs to have both the switch and the NIC explicitly set to the proper speed and duplex setting (like 100FD).


Pete

[Chris Vail]

The best deal is Glance Plus: it gives all kinds of cool graphs and stuff. This is an optional additional piece of software you can order online or get a codeword from your reseller.

You can get much of the same thing from sar or netstat. There's all kinds of threads on these forums concerning this. Just search for a few of them for an education.


Chris

[Uday_S_Ankolekar]

Use Glance or Openview Reporter to findout the Network utilization .
netstat -rn will also give you few info
Also check the speed setup on each NIC if it's configured to full duplex. You can check this with lanadmin command

-USA..

[Tim D Fulford]

You can set up nettl to log everything, then use netfmt to debug it. That is if you want to go & see the N'th detail.

My preference is MeasureWare (NETIF_xxx parameters) which can look at network traffic by interface. Then glancePlus & lanscan

Regards

Tim

[John Bolene]

The best way to improve network performance is to install a GigE interface in each machine and direct connect them with a crossover cable.

We have several Oracle frontend and database servers connected this way.

[Steven E. Protter]

I like the Gigabit idea. Especially since most new switches will handle it.

Before you spend the money, measure the performance. Glance is pretty, but sar scripts are better tools.

I am attaching a good one. It collects data background and can be configured to run for any time period you wish to get good averages.

SEP

[Caesar_3]

Hello!

Use the glance plus, it will show you
all bottelneck with the network.
Also you can install the tcpdump from
which you could read a lot info about
what go on the net and how much.
http://hpux.connect.org.uk/

Caesar

[Ron Kinner]

IF you have several routers along the way then you might want to look at MRTG. It also draws nice graphs of the traffic in and out on each interface of the routers and PCs and can also graph things like dropped queue packets and errors if you want it to.

It's free and runs on UNIX or NT.
http://mrtg.hdl.com/mrtg.html

Great for finding bottlenecks.

Ron

[Jeffrey Killian · Accepted solution]

We too run Psoft with Tuxedo Application servers and Process Schedulers (batch server). Sounds like you have appservers/batchservers/databaseserver running on separate machines?

How many network interfaces do you have available in each of these machines? Creating a separate network between these machines alone will give you good improvements at VERY low cost. We have come a long way in our network environment and would be happy to share. Have you even considered running the appservers on the same machine that runs the database - we evolved to this and the preformance is awesome...

Have you done tuning of the SQL that is accessing the database - we acheived incredible results by doing this -

[Edgar Avila]

Use NTOP
It's GPL. It means, almost free.
You can view the amount of traffic betwen servers, the top bw, top multicast, etc.
Last version has a wonderful web interfase.

Most of the time bottlenecks it's a collision issue. You can determine which it's generating the noise viewing the ammount of traffic generated. Then check the interfase of that server.

If you're using your main interfase to do backups then maybe it's time to add another NIC for that.