I am curious about others network deployment manager architectures.  I have never installed dman on a dedicated dman only box, but I do see a small benefit. I also see some overhead. In my opinion, the cons outweigh the pros.

To be clear, the following scenario is my question.

A cell consisting of 3 boxes.
Box 1: dedicated deployment manager only.
Box 2: Node 1 (app server box).
Box 3: Node2. (app server box).

Does anyone use this? I dont think its wrong really, just a waste of resources and time.

To further complicate things, has anyone ever implemented the following:

Two cells consisting of 5 boxes:
Box 1: dedicated dman box (dmgr for cell 1 and dmgr for cell2)
Box 2: Node 1 Cell 1
Box 3: Node 2 Cell 1
Box 4: Node 1 Cell 2
Box 5: Node 2 Cell 2

The point here is more than one dmgr on the same box. I see no good reason for this whatsoever. Specifically, is this approach recommened or not recommended? If this is not recommended, can someone point me to a white paper or Redbook or something that speifically points out this is a bad approach? I really need it. Thanks.

Hi Dan,
 
 I use you first approch more than one time (v6.1, v7 and v8) with very good result, and previsuly installed cells (DMGR in one of the nodes) migrated to this new topology virtualizing DMGR machine.
 
 For executting jython scripts I have found very good (and more faster) the use of Admin Thin Client instead of the traditional wsadmin from DMGR machine.
 
 As you say the overhead of the DMGR in the same node impact the performance.
 
 Important to take in mind, with this approach you have three machines to license instead of two. The numbers are important too :-)
 
 Your last suggestion is possible (DMGRs must be in differents ports

 About your references i haven't found nothing but maybe there.

  Hope this helps,

regards

Thank You. Yes I know #2 is possible and yes I know you need to change ports. Its not clear to me just why anyone would want to do this and I wondered if anyone does or not.  It seems like a large waste.  Thanks for the reply.

Dan -

I actually run a Shared Services Infrastructure, and have for a few years now, that is similar to the question your're asking.  The reason I do this is because with the advancements in hardware, most of a server sits idle in the rack compared to what teams actually need.  

By sharing the harware, I can drive denisty on a server and achieve higher utilization.  I reduce needed rack space, power and cooling costs decrease consequently, as does licensing cost as my footprint is smaller.

As for the topology, I tend to deploy my Deployment Managers in there own VM.  I benefit from the inherit benefits of virtualization, and by seperating the VMs, it allows maintenance on the DMGR of any system and leaves the other DMGRs up.

An important consider you would need to make is whether you want all of your applications to have their own cells.  Granted, if you're running WAS 7.0 & WAS 8.0 you should be doing this, but if you have multiple applciations that are WAS v8, why not use a single DMGR for both and combine the cells?

Assuming you are running multiple versions of the same version of WAS, you can install multiple binaries of WAS on the same host.  This allows you to run your apps independant of each other (including maintenance, runtime environments, etc.)  There's REAL benefit to doing this.  And if the binaries are on the same host, and federated to the same DMGR in the same cell, you avoid the port issue.

I digress though...

The real reason I do this really comes down to what I mentioned earlier - COST. It is significantly more cost effective to do this.

Let us know if you have more questions about why a company would do this.

Erik