Xiaodong Yi wrote: > It is not a typo. I copied from UnixBench output directly. Howver, it > must be a bug of Luvalley because even the native Linux benchmark on > Double-Precision Whetstone is not that high. I also noticed that other > benchmarks are all lower than native Linux. > > About timing, Luvalley does nothing more than KVM except that Luvalley > implemented the VLAPIC timer using TSC while KVM uses the services of > Linux kernel. The other timers of both Luvalley and KVM, I think, are > all implemented in Qemu. > > Moreover, I could not explain why Luvalley's benchmarks on process > creation, execl throughput, file copy and shell script are 20% ~ 40% > higher than KVM, while other benchmarks on pipe throughput, context > switching and syscall overhead are almost the same as KVM. > A typical issue in VMM benchmarking using OS benchmark such as what you used is time inaccurate issue. Benchmarks using guest time won't be able to get a right time or right duration due to scheduler etc, and thus VMM benchmarks is using network time for measuring such as vConsolidate. Spec.org is definning their benchmark for VMM, and I believe they will use network time too. For simplicity, you may continue use OS benchmark to measure VMM, but then you need to calibrate guest time accuracy first such as using stop watch etc. In both Xen & KVM, we benchmark using OS benchmark too, but it is usually only to see improvement of a performance patch. Formal benchmark data needs to consult wall clock or stop watch. Thx, eddie-- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
