On Sat, May 2, 2009 at 12:46 PM, Phil Smith III <li...@akphs.com> wrote:
> The pathological cases of "This guest runs away, real Swap DASD gives me a > chance to control it" are perfectly valid, but don't contradict the statement > that VDISK is better *for performance*; rather, they support it. Let me show you the numbers on a beer coaster then... I worked with Terry on his setup and recommended that in this exceptional case it would make sense to "put a gravel bed next to the road to slow down the penguin with no brakes" It turns out their application had a bug/failure in that it had apparently an endless demand for memory in some situation. Since VDISK is fast, you can fill it up pretty quick - at 40 MB/s your 2G VDISK is full in a minute. But filling up the VDISK did not hurt z/VM, it just made Linux give up when it ran out of swap space. And that hurt their application (probably the ungraceful shutdown required journal recovery etc of a large disk, etc). If you do the math, you will see that a penguin with intention to kill itself could also fill a 2G real disk in 10 minutes or so. In this case that was enough because they were sitting next to it during the test. When you don't expect it to happen, you will not be there in time to rescue the penguin. So I conclude: when the memory requirements of the application exceed your planned capacity "just a little" (say less than the swap space you set up to support Linux memory over commit) then VDISK will do fine. It does not hurt z/VM and it avoids people complain about the sudden slowdown. When Linux needs way more than planned, it will fill up swap space (whether VDISK or real) and the Out-of-Memory Killer will stop vital processes during the crash. The only advantage of the real swap disk is that it postpones the moment of crash with an hour and costs more money. Not a hard choice for me. Tuning by opinion and fear rarely is cost-effective. Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/