<html><head></head><body dir="auto" style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;"><br><blockquote type="cite">On Jun 20, 2016, at 13:25, Paul Mackerras <paulus@ozlabs.org> wrote:<br><br>On Mon, May 16, 2016 at 02:58:18PM +0800, Li Zhong wrote:<br><blockquote type="cite">This patch tries to use smaller locks for each irq in the ics, instead<br>of a lock at the ics level, to provide better scalability.<br></blockquote><br>This looks like a worth-while thing to do. Do you have any<br>performance measurements to justify the change? This will increase<br>the size of struct kvmppc_ics by 4kB, so it would be useful to show<br>the performance increase that justifies it.<br></blockquote><br>Actually, I saw some “improvement” because of the vcpus were not binded, io jobs and irqs on the guest were not binded. After I fixed those random factors, the result became stable, but I couldn’t see any obvious improvements from the patches... <br><br>Maybe I need find some other test cases that could support this change. <br><br><blockquote type="cite"><br>Also, when you resend the patch, please make the patch description<br>more definite - say "With this patch, we use" rather than "this patch<br>tries to use", for instance.<br></blockquote><br>OK, I will change that when doing a resend, if I can find some workload that could benefit from this change. <br><br>Thanks, Zhong<br><br><blockquote type="cite"><br>Regards,<br>Paul.<br>_______________________________________________<br>Linuxppc-dev mailing list<br>Linuxppc-dev@lists.ozlabs.org<br>https://lists.ozlabs.org/listinfo/linuxppc-dev<br></blockquote><br></body></html>