TruCluster
Showing results for 
Search instead for 
Do you mean 

Tru64 5.1B3 Cluster slow with 2nd member added

Occasional Advisor

Tru64 5.1B3 Cluster slow with 2nd member added

Hi !

I have set up a tru64 5.1B3 Cluster with 2x DS20 (1G + 3,5G RAM) using EVA3000 storage.
Cluster interconnect is via cross-over cable 100MBit (both NICs set to 100FD via console).
When the cluster was set up, there was initially only one DS20(1G RAM). All applications were performing well and users were happy.
After 2 weeks, I added the second DS20 (3.5G RAM) and the users started to complain about slow performance. The main application is a terminal based software making heavy use of file locking. When both systems are online, users are logged in to both systems via cluster-alias.
I expect CPU and Mem is not an issue as everything is running well on one system.
I observe up to 3000 packets on CI interface during operation (using netstat and monitor).

I now removed the second DS20 and everything is running on the first DS20 again with good performance.

Any hints welcome.

Michael
2 REPLIES
Honored Contributor

Re: Tru64 5.1B3 Cluster slow with 2nd member added

First of all... may we assume that all shared devices were properly visible ot each node (presented to both node). The Tru64 Clsuter software is so good it will hide issues here and start (DRD) serving non-directly-connected disks over the CI.

next, the CFS (Cluster File System) on top of the AdvFS on disk structure make 1 node in the cluster completetly responsible for new ALLOCATIONS for a given mount point. Check out 'man csfmgr'.
That may well cause a bottleneck.

>> Cluster interconnect is via cross-over cable 100MBit (both NICs set to 100FD via conole).

That's not too great, and may well have been the bottleneck.

Invest the extra few $$$ and get GB nics and a little GB switch. I'm running GB here at home for a hobby/test network between laptop, DS10 and RX2620 using a $35 8 port D-link auto-everything hub. Surely your production site is worth that!


Hope this helps enough,

Contact me if interested in one-on-one for fee help.

Hein van den Heuvel
HvdH Performance Consulting
Honored Contributor

Re: Tru64 5.1B3 Cluster slow with 2nd member added

This is very tipical of lan interconnect when the cluster interconnect is not working correctly. I use to have these problems when the cluster interconnect is gigabit ethernet and the autonegotiation cannot be disabled.

You must work in troubleshooting your cluster interconnect, try modifying the network interface parameters at console level and using the lan_config command. Place the right lan_config command in /etc/inet.local.

Also, ensure that your file systems are served by the right node. Use cfsmgr to display who owns a specific filesystem and relocate them if neccesary.
Por que hacerlo dificil si es posible hacerlo facil? - Why do it the hard way, when you can do it the easy way?
//Add this to "OnDomLoad" event