Wim has some good suggestions, but I fear you're not ready for that level of sophistication yet. So I'll start back at the beginning.
The way I most often see this set up is with one node (with the fastest CPU that a given site can afford to have an Analyzer license from CA for) to do the analysis; all the other nodes run the Collector. (When you set up the Data Collector in a cluster, create one common PSDC$DATABASE area for all the nodes in that cluster, rather than a separate area for each node.)
On the Analyzer node, add all the other nodes to its Collector database with
$ADVISE COLLECT MODIFY CPD/NODE=name
commands. Then all the nodes will show up in the GUI for selection.
Next you'll either need to copy all the .CPD files to the Analyzer node's PSDC$DATABASE area nightly (sometime after midnight, which is when the old .CPD file is closed and a new .CPD file is started for the new day), or else just point to them in-place over DECnet from the Analyzer node:
If you want to avoid actually copying all the .CPD files, then on the Analyzer node, define the PSDC$DATABASE logical name as a search list that includes your local PSDC$DATABASE: area (for the database of nodenames you just expanded to include all nodes) plus the DECnet-style network directory specifications for all the different PSDC$DATABASE: areas for all the remote nodes or clusters. (This is best done with DECnet proxies set up, so you don't have username/passwords in your DCL files or in the logical name definition.) (And if you have a DECnet cluster alias for a cluster, you need only point to the PSDC$DATABASE area for that cluster once, using the alias, to gain access for all the .CPD files for that cluster.) This definition of the PSDC$DATABASE logical name can be a process logical name -- it doesn't have to be system-wide.
To get started, I'd recommend using the DECwindows/Motif GUI interface, which you invoke using
$ADVISE PERF/INTERFACE=MOTIF
If your analysis node isn't a VMS workstation or server with a display monitor, you can use a $SET DISPLAY command just before invoking $ADVISE to send the display to another node with X-windows display capability (I do this with eXcursion software on my Windows PC, but Hummingbird Exceed or Reflections X would also work).
Start by selecting a set of nodes and a time period for analysis. (Do one cluster at a time.) It will take a bit of time for the Analyzer to do the number-crunching. After the number-crunching is done, you can look at the graphs and reports. To get the quickest return on my investment of time, I usually start with the Brief Analysis Report, drilling down for anything that looks interesting -- it will catch most of the low-hanging fruit (easy things to fix). Next, I usually look at most of the graphs, looking for things that look unusual (and the definition of "unusual" is something that, unfortunately, one must learn with time and experience).
One unfortunate aspect of this tool is that you have to set the time period at the time you select the data, so if you want to zoom in or look at a different time period, you have to re-do the calculations.
Once you get a feel for what information is useful, you may want to do more automation. The tool is set up well for this, with DCL command equivalents for basically anything you can do with the GUI. For example, you might calculate the graph data each night in a batch job and save it in a file for loading and viewing later from the GUI, allowing you to skip the time-consuming number-crunching step, or you might just create all the possible graphs at once, as Wim has suggested (and all the reports can likewise be generated at the same time).
Note that when you install the Data Collector it defaults to deleting .CPD files after one week, in the interest of not filling up all available disk space, so you'll probably want to do something about that: either make sure you extend the retention time, or else set up a mechanism to archive the .CPD files somewhere safe (to tape, or to another disk or directory) before they get deleted automatically.
