On Mon, Oct 20, 2008 at 6:04 PM, Ron Foster at Baldor-IS <[EMAIL PROTECTED]> wrote:
> setting kernel.hz_timer when you are running SLES10 in an LPAR. Don't bother. The main reason is to ensure that idle Linux guests drop from queue so z/VM can do memory management. But in LPAR you dedicate memory to that server, so there's nothing to manage. And when you run a recent SAP it comes with Java code and other stuff that prevents the server from going truly idle anyway. In its first incarnation (SLES8, I think) the "on-demand timer" could in some cases cause extra overhead for busy systems. That extra overhead does not exist anymore with the current code. I don't think we ^H^H anyone has measured it, but have advocated the default of "1" based on "if it's cheaper it must be slower" PS I'm biting my tongue regarding the motivation to run LPARs because of z/VM hiccups. If z/VM is unstable then you should work with IBM to get that fixed. If you're having unexplainable performance issues, then get a performance monitor and someone who can help you understand the numbers. Moving a major part of the workload to LPAR is making that job harder because you have less options to measure there. Rob -- Rob van der Heij Velocity Software http://www.velocitysoftware.com/ ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
