On 6/14/07, Mark Post <[EMAIL PROTECTED]> wrote:
Unless those guests are pegged at 100% CPU for that one CPU, giving them a second virtual CPU won't help, and will likely hurt. According to Rob van der Heij, adding a second CPU frequently causes the guest to not drop out of queue, and the z/VM scheduler treats it as a batch workload instead of interactive.
"When in doubt, one virtual CPU will do. When you're know better, probably too." :-) Guessing about performance is something I only do for recreational purposes. Things are rarely obvious and sometimes it helps to explain why your tuning my have the opposite effect. The case that you refer to is the interaction between CP and Linux scheduling with a virtual MP, which makes the virtual machine end up in Q3. But for Linux to do I/O you need it to be dispatched quick and frequently to reflect the I/O interrupt and let it do another SSCH. That does not work well in Q3. And of course it gets even worse when the benchmark itself also does the measurements and wishes to implement timing by busy wait... :-( Rob ---------------------------------------------------------------------- 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
