Originally posted by: spigven

My HACMP and IBM Hardware skills are a decade out of date but soon I will have a 2 site 3 node cluster to configure, all P series. Site A will be production site with 2 nodes and shared storage via a couple of SAN switches and a DS5020. Site B will be single node for DR purposes and storage will be directly connected to node B. I'm ploughing through the on-line documentation currently but would appreciate someone else's opinion. First question is the order of work, I was thinking:

Configure site A first as a 2 node single site cluster with non concurrent vg's and mutual takeover.
Add in site B. Remote data mirroring to site B using GLVM in asynch mode across IP WAN. So is this a valid approach or does it have to be multi site from the off ?

Not sure about the site to site heartbeat, maybe use the XD WANs ?

Also not sure about the system console(s). Virtualisation is not going to be used, for configuration purposes can a green screen (wyse ?) terminal and kbd be attached via serial port ?
#AIX-Forum

Originally posted by: afisher

spigven,

I'm currently in the middle of building a 2-site 2-node XD cluster for a customer. A couple things to consider, do you have Layer2 from site to site? If so, I assume you're planning to use IPAT via aliasing. Please say no HWAT, right?

I see you're opting for async GLVM. What's your inter-site link size? Don't have the bandwidth to do synchronous GLVM? Or what was thought here?

If you're doing a 3-node cluster, you could setup a triple-mirror RPV relationship between nodes.

Yes, you can use terminals... it's not pretty, but yes, you can.

HA will whine and complain that you don't have an alternative heartbeat to ethernet. It actually recommends you use Serial heartbeats, as if you have a serial cable that stretches between your sites. To bridge this gap, I've used Perle IO/Lan converters to spoof serial heartbeats and just run it over a secluded ethernet link between sites (if possible). They're cheap little boxes and come with a built-in GUI for support and config. http://www.perle.com/products/IOLAN-DS-Terminal-Server.shtml

I have complete HACMP/XD walk throughs on my site for free if interested. Granted, it's using 5.3 but they still apply.

-Drew
www.NotesOfANerd.com
(A Public Knowledge Repository for UNIX/Storage/Clustering/and more)
#AIX-Forum

Originally posted by: spigven

At the moment I only have scant details and I'm not currently directly involved in the planning although I'm pushing to be more involved.

Its supposed to be a high availability solution with DR, someone has produced a 2 site 2 node diagram with 'system mirror' linking the 2 sites. To me that is a DR solution not HA with DR which is why I've suggested a second node at the main site.

The only cluster I've been involved with used IPAT and HWAT and I was expecting to do that ?

By my calculation each node will need 7 nics(including 2 for RPV) and 3 subnets.

Dont know yet how far apart the sites are or what the WAN will be. Gravitated toward asynch to be on the safe side, don't want to impact performance on the main site.

For GLVM the 2 nodes at main site will each host an app and have mutual take over of some form. For GLVM the relationship will be initiators on each node at main site to responders at single node at DR site (plus same again in the opposite direction). I'm about to do some studying on the GLVM front but do they become part of the resource groups and is this an appropriate solution ?

As there is thankfully no Virtualisation involved I saw an HMC as an added complication, at least initially and was not planning to use GUI in the config process.

I will investigate advice re serial heartbeat and your website.

Advice much appreciated.

Mark.
#AIX-Forum

Originally posted by: Casey_B

The one reason that you would want to consider IPAT via aliasing is you need less adapters.

The other is that Hardware Address Takeover is typically slower, and not necessary with
modern networking hardware.

There is a provision with HA/XD to use a set of adapters in an XD_ip network, and those can be used
for heartbeating.

There is also an XD_rs232 network type, but I would assume that in most cases, this is not practical.

If you are using async GLVM, remember that it is in practice a write caching system.
So...Your data is written locally before it is sent across the wire, and your application is told that the
write succeeded.

This means that if you have some data stored locally, that has not been transferred, and you have a
disaster recovery, you need to consider the possibility of data divergence.

One last thing...There is also a seperate PowerHA forum on developer works.

Good luck, Hope this helps,
Casey
#AIX-Forum