Securepath installed with only one HBA - issues?

[Richard Adams_7]

Are there any known issues with running securepath on a box with only a single HBA, particularly when MS clustering is involved?

I have a couple of DL380 G3s with a single HBA and securepath installed on each.(Reason for this is that I'm using a 'live' base server build for boxes in my testlab where we can't justify the expense of second HBAs). I'm showing a single lun on an HSG80 to both servers and attempting to run MS clustering on same. While clustering appears to work fine, any attempt to failover disk resources fails. As soon as I remove securepath everything works correctly as I would expect.

I run a large number of MS clusters in my live environment with a very similar config - the only difference is that they all have 2 HBAs and therefore have no issues with securepath.

My question is, while you would obviously not usually install securepath on a single HBA box does it cause any problems if you do?

[Fernando Bandeira · Accepted solution]

Hi Richard,

If your HSG80 storage system is running in Multi-bus failover, it is actually required for you to install Secure Path on your server even if you only have one HBA. The reason is that each controller has two host ports, and there are two controllers, so a HBA at any time can see two host ports (thats why there are two connections for each HBA). So if you dont install Secure Path, you have an HBA which can see two active ports, and therefore two ways of connecting to your LUNs, and no way of the host understanding/controlling this.

Regards,

Fernando

[Richard Adams_7]

Thanks for that - I'll do some more digging from the clustering angle.

[Uwe Zessin]

Whether a failover works or not is heavily dependend on the way the connections from the fibre channel adapter to a controller ports run.

In bultibus-failover mode, the HSG80 presents a unit through all 4 controller ports, but I/O can only be done throungh the controller on which the unit is 'online'.

[Dana Lord]

You would need 2 LUNS, one for mscs quorum and one for your data. LUN's must also have same logical assignments in Windows and both HBA's need access rights to LUN's at controller.

[Richard Adams_7]

Some further findings now I have found time for some more testing. Remember, I am ONLY experiencing the problems in my test environment where due to cost constraints I only have single HBAs available per server, so the discussion is rather academic. My live (2 HBA per server) setup works fine as expected.

As an experiment I reinstalled securepath on both of my two two-node clusters. Straight away it became impossible to fail resources between nodes. Nothing of any relevance in the cluster or event logs. Removing securepath restored normal failover capabilities at once!