On Dec 7, 2007 4:23 PM, Patrick Spinler <[EMAIL PROTECTED]> wrote: > In brief: when load balancing client network connections across > multiple instances of an application, is it better to run multiple > separate 1 virtual IFL linux guests, each with a separate copy of the > application, or a single, large N virtual IFL linux guest with several > copies of the application running inside it?
Load balancing can be an issue. When using techniques from the discrete world (like "sprayer" or "round robin") you end up with multiple virtual machines kept warm. This may prevent CP taking resources away from idle virtual machines. When possible, consider something that "overflows" to the next free virtual machine in a stack. > Assumption: you have enough real IFL's to back the number of virtual > IFL's you'd assign. Rather than the real number of CPUs you should look at the chances the virtual machine can get al its CPUs dispatched at the same time. If contention is high enough, adding more virtual CPUs will make it worse for yourself and others. > With expansion we have budgeted for next year, though, we'll have > enough real IFL's to consider consolidating all these TEMS proceses > into a single, multi-virtual-ifl linux guess. There's several issues with "virtual-MP" Linux guests that affect scalability. The only reason that would make me prefer scaling vertically is when the "idle load" of your application is so high that it outweighs the extra cost of virtual-MP. Your performance monitor would be your friend to measure the various scenarios and make a more educated decision. > We're not sure if there would be any cp scheduling benefits to either > configuration, or any resource benefits (e.g. storage usage, or paging). There's also things like security, ease of management, accounting and charge back, service levels, administrator tasks, priority, etc. Some of which may or may not apply to your scenario. > Help support our fluid fueled debates ! What're the group's thoughts, > please? Discussing performance problems over a beer is a good approach ;-) But you should also spend some time on measurements... Rob -- Rob van der Heij Velocity Software, Inc http://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
