On Wed, Jul 29, 2009 at 9:52 AM, Berry van Sleeuwen<[email protected]> wrote:
> We have received a fixtest from IBM for testing but we need to recreate the > problem on our test environment. So far we did not succeed in recreating the > problem. As it turns out we need to get one or more users into the dispach > queue without using CPU cycles. > > The problem on our production systems looks like to be caused by linux > guests that are a bit misbehaving on VM. Even while they hardly use CPU they > remain in the dispatch queue. My guess would be because some tools keep the > machine busy with query requests. Right, that is a very familiar problem with a lot of middleware and enterprise applications. It's called "polling" when the developer writes the program to wake up very frequent just to check for work. It is bad behavior and hurts scalability. > Our test environment doesn't have any guests that remain dispatched at all > time. So I need to get a user, either CMS or VSE, to wakeup as often as to > force the guest to remain in the dispatch queue. But the catch is that the > guest should not use CPU, so a looping user won't work here. Any idea's how > to do this? Try this: PIPE literal +0.1 | dup * | delay | hole On a z9 it shows me using less than 0.01% of a CPU and sit in Q3 all the time. If this is not good enough, call again.. I am working with a customer on a problem where servers remain in queue when they should not. Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/
