Network capacity & performance

[Paul Condren]

Hi.

We have an oldish server using 100base=t cards but the switches around it have since been upgraded to 1000 base-t switches.

The servers are getting some memory and CPU upgrades shortly and ive suggested that the cards should also be changed to faster 1000 cards.

However Ive now been asked to demonstrate that there would be a performace increse from the new cards.

The bulk of the network traffic to this server is large batch file transfers to and from other syatems, and we have used APA to group 2 100 speed cards together to help the performance.

Netstat etc wont show me what %age utilized the current cards are. Nor will glance. Is there anything else I can use to show that at peak times this cards are a bottleneck?

Searching the net suggested network node manager but I think this is chargeable.

thanks.

[Pete Randall]

Paul,

Yes, NNM is chargeable. Did you look at gpm > Reports > Network Info > Network By Interface ? That particular report gives you some real numbers that you can use to compare idle vs load.


Pete

[Tim Nelson]

Do a search of this forum for lan statistics.

One relevant item is here:
http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1199612

Yes NNM is a purchasable item. $50k - $100k

Glance is a great tool, actually a must have for HPUX.

netstat and lanadmin can also be used to gather usage stats.

Of course you will see much greater performance with GigEther but only if you are stressing your current 200MB aggregate. If you only push/pull low volumes then why waste the money.

Gather some stats over time, if you are anywhere near 100MB or 200MB / second then show the stats to the boss.

With the upgrade of the Gig switches make sure you check on your handshaking. AUTO never worked well with 100MB nics. Lock then to 100FD.