back to back cluster

[Edgar Ulloa]

I have a lan cluster, Hp give me a suggestion, connect my nodes back to back (card- cross cable - card)

when I connect the nodes phisicly, what I need to do for my cluster know the new connection..?

the connection will be detect automacly..?

how I configure tcp or decnet IV for this conection..?

any command or comment..?

Thanks for your help

[Mike Reznak]

Hi,

cluster concepts.

http://h71000.www7.hp.com/doc/82FINAL/6318/6318PRO.HTML

or folders for older version of OpenVMS.

http://h71000.www7.hp.com/doc/os82_index.html

Mike

[Willem Grooters]

The principle is almost too simple:
1. Load the required licenses on both machines (don't forget the CLUSTER license)
2. Make the physical connection.
3. Run the CLUSTER$CONFIG procedure on one machine. Here, you CREATE the cluster, giving it a name, number and password.
4. Reboot machine # 1
5. Run the CLUSTER$CONFIG procedure on the other machine and JOIN the cluster. Here, you specify number and password of the cluster to join.
6. Reboot machine # 2

and you're done.

And since it's VMS (well designed) there is no need for DECNet nor TCPIP (Cluster communication uses a separate protocol).

There are quite a number of things to consider if you want to take advantage of all facilities a VMS cluster offers (there is a whole manual about cluster configurations), but this is the base of all. Once it took me about 15 minutes ;-)

[David Jones_21]

I'd only do this if both machines have 2 more ethernet adapters and then configure all non-cluster protocols only on the other adapters. Otherwise it gets messy because 1 machine would have to act as a router for the other.

Considering the time it takes some managed ethernet switches to boot, having a direct ethernet connection for LAVC can make the cluster more robust. The machines shouldn't have any problem finding themselves over the dedicated link.

[Edgar Ulloa]

Its necesary assembly again the cluster..?

right now the cluster is assembled and runing, but its only lan cluster.

If I wan connect back to back for hardbit I need rebuild again with cluster_configure?

Thanks for your comments

[David Jones_21]

VMS configures all ethernets it sees for cluster traffic by default, so you don't need to redo cluster_config to get it to work.

[Jan van den Ende]

Edgar,

so, you HAVE an existing cluster:

- Shutdown
- do the physical instal
- reboot

The cluster will automagically use all available interconnects.

(if you would venture and succeed in installing the new connection in the RUNNING cluster, even the reboot would not be necessary! )

Proost.

Have one on me.

jpe

[Ian Miller.]

If the two nodes already use the LAN for cluster traffic then there is no reconfiguration required.

Are the nodes connected to a network switch/hub/other and you plan to replace this with a simple cable?

[Bojan Nemec]

Edgar,

If I understand yours situation, you have two configured cluster nodes in lan cluster. Each node has two or more lan adapters. One of these adapters on each node is free (not connected).

If this is correct just connect the cross cable betwen the adapters and you are done. The connection will be detected automaticaly.

Bojan

[Andy Bustamante]

Edgar,

Bojan is correct, if you have open network ports you just need to connect them. The cluster will automatically begin to use the new path for SCS (cluster) traffic.

You do need to make sure the ports are both configured for the same speed and duplex settings. You can use the LANCP utility for this.

$ MCR LANCP SHOW DEVICE
$ MCR LANCP SET DEVICE EIB0/SPEED=100/FULL_DUPLEX
$ MCR LANCP DEFINE DEVICE EIB0/SPEED=100/FULL_DUPLEX

Note for your networking team, you don't want a second VLAN on the same hardware the primary network interfaces are connected to. This is a redundant path to ensure cluster availability. For two nodes a cross over cable will do the job. For more than two nodes, you need to have a redundant network connection, either a switch or hub.

Andy

[Edgar Ulloa]

Thanks for all your comments
I will the conection early

my doubth was because when I put
sda>sho port only I can see one port

866B7E30 pe PEA0 SYS$PEDRIVER
and I want have 2 ports betwen the nodes.

I think that will be necesary configure independiently each card or some thing.

Thanks again

[David Jones_21]

>>my doubth was because when I put
sda>sho port only I can see one port

866B7E30 pe PEA0 SYS$PEDRIVER
and I want have 2 ports betwen the nodes.

I think that will be necesary configure independiently each card or some thing<<

Clustering began using a special piece of 'CI' (Computer Interconnect) hardware. When clustering via a LAN, the PEDRIVER (Port Emulator) provides a pseudo-device that looks like a CI but communicates over other network interfaces. The code inside PEDRIVER handles selecting the path to use so you only need PEA0.

The DCL command "MCR LANCP show config/users" will show you which protocols are being used the available ethernet adapters. Cluster traffic is carried on NISCA.

[Edgar Ulloa]

Ian Miller

You asked me..?

Are the nodes connected to a network switch/hub/other and you plan to replace this with a simple cable?

the idea is have both, switch and simple cable..? its probably do it..?

David Jones

Thanks for your comements Can I select wich card I choice for my cluster trafic..?

Thanks

[David Jones_21]

>>Thanks for your comements Can I select which card I choice for my cluster trafic..?<<

The sys$examples directory has 2 programs, lavc$start_bus and lavc$stop_bus, that control which controllers run NISCS. See appendices D and E of the "Cluster Systems" manual.

I stop LAVC on my fastest server's gigabit card (keeping it on the 2 100 Mbit ethernets on the motherboard) because otherwise it tends to overwhelm the slower Alphas that have only 100 Mbit interfaces.

[Ian Miller. · Accepted solution]

on VMS V7.3 and later use SCACP to control which interface is used.
http://h71000.www7.hp.com/doc/82FINAL/6048/6048pro_contents_003.html#toc_chapter_18

[comarow]

You don't say if you have more than one cluster on your network. Make sure the cluster has a unique group number and group password.

This is during cluster config, but can also be performed using the sysman utility.
The defualt is group number of one and it can get you into a lot of trouble.

You can take a cluster_authorization.dat
from one node, sys$common:[sysexe] and copy it to the other nodes.

Bob

[Edgar Ulloa]

Thanks for all your comments

I will apply it