On 7/25/09 2:46 AM, "Rob van der Heij" <[email protected]> wrote:
> On Sat, Jul 25, 2009 at 3:59 AM, David Boyes<[email protected]> wrote: > >> If you want this to work, then you need to configure CSE in your VM >> installation so that CP enforces that a user can be logged in on only one >> system at a time. You'll need RSCS (or an equivalent NJE implementation) and >> PVM to make this completely safe. > > RSCS and PVM don't add any safety to CSE beyond XLINK. XLINK is > enough. Many shops have done this. Um, what happens if/when a directory entry gets out of sync? Yes, you can invent a solution, but is the cost of maintaining something external that duplicates already available *and supported by IBM* function worth it? How do you handle scaling up the environment and monitoring tools? Who's paying for supporting your local hacks? Whose head rolls if/when it fails? > Suggesting that people would need to get special bids for PVM and > license RSCS just scares them off and motivates them to walk around > with a loaded gun. I don't think I agree, and it's probably a worldview difference. IMHO, people are the expensive resource, and it's people that make mistakes. Clusters add opportunities to make mistakes linear to the number of nodes involved, and the impact of mistakes is exponential in terms of probability of causing the entire cluster to fail. The computers do what they are told, even if it's stupid. Any opportunity to take people out of the loop in terms of managing and maintaining the system adds to safety. So, I think that the exhortation to use CSE *as it was designed to be used* is actually a pretty good idea. You can certainly design additional semantics on top of the fully implemented CSE, but you're starting at a more advanced place with a richer toolkit, and the full CSE is a very helpful thing for implementing a reliable cluster. So, we can agree to differ. *I* think that starting with a fully-enabled CSE is a safer option in terms of having the SYSTEM manage itself and prevent at least two classes of possible stupid actions, and *I* consider having that extra safety a good thing, especially for a newbie who doesn't have the VM skills to implement the replacements. YMMV. -- db
