Where to run Data Server of Availability manager v3.0 ?

[Jan van den Boogaard_1]

Hi all,

We think of using Availability manager v3.0 because of the WAN functionality.
Using Avail.mgr V3 means that we need at least 1 data server for each extended-LAN.

What would be better/wiser/nicer, running the data server on a dedicated windows box or on a dedicated VMS box? or even in the cluster itself?

Running 1 windows box per LAN would mean a single-point-of-failure?

[Richard Brodie_1]

Running it on the cluster you are monitoring is definitely a bad idea. If the cluster locks up then the data server will too, which somewhat defeats the point of having the low-level collector backdoor into the system.

I would probably go for dedicated windows boxes for monitoring. It depends what level of support/how far away the remote sites are how available they need to be.

[Ian Miller.]

Run it on two small systems - vms or windows depending on what you have available. Getting the budget for system monitoring systems is always a problem :-(

You could run it on the cluster and on a separate box.

[Jan van den Boogaard_1]

Thanx Richard and Ian

Running a data server on the cluster to be monitored wouldnt work.

1 other question.
Suppose there is a cluster across several locations, e.g. 2 locations.

Then, to be disaster tolerant, a windows box or small vms box with a data server is needed on each location.

Suppose there are several clusters, spread over 2 or more locations. Then each cluster could be the Data server on VMS for the other clusters. or would you still need separate boxes then?

[Ian Miller.]

I think one monitoring system running the data server per site would work.

Perhaps it should be two per site for resilience. Could be a small windows box, a small vms system (which could be running other things), or a small linux box (possible but unsupported)

[Duncan Morris · Accepted solution]

Jan,

we have several clusters spread over 2 locations. These have a common GigE bacbone and I have configured data servers at each location for redundancy.

If you were to use cluster A to monitor cluster B, then you need to consider the problem where failure of the inter-site link, or of one location, might well cause all of the clusters at the other site to hang for lack of quorum.

I would therefore recommend use of a standalone box as the data server. Monitoring a combination of 23 node, 12 node, and 6 node clusters on a 617MHz DS10 uses remarkably few resources.

Duncan

[Jan van den Boogaard_1]

Duncan and Ian, thanks for yr replies!

To avoid any situation that there is no data server on a cluster available because a number of clusters or even all clusters are in a hang, dedicated small alphas or windows boxes are better.

[John Gillings]

Jan,

We have several clusters split between two sites, also with multiple networks for various production, and non-production systems. There are two small systems (DS10s), one at each site, running OpenVMS which are multi homed, seeing most of the networks. Each is configured as an Availability Manager server. We can run the monitor on our desktop PCs (non-production network) and see production systems.

Before V3, we ran the Java AM application on the same DS10s. Displaying a dozen or so systems consumed a minimum of 50% of the CPU and anything up to 80%! The new model is much less CPU intensive, and far more flexible (as we can now have multiple monitors).

If you have a dedicated box, costs would probably push it towards using a PC as your server. As long as it doesn't do anything else, there probably aren't any compelling reasons to prefer OpenVMS over a PC as your server.

[Barry Kierstein II]

One item to note: If you need to connect to multiple networks via multiple adapters, the Windows Data Server will do that. The Windows driver can open multiple adapters at the same time. This is described in section 6.3.2 in the AM Installation Instructions.