Date: Fri, 31 Jan 2014 19:48:48 -0800 From: Peter Grehan <grehan@freebsd.org> To: Andrea Brancatelli <abrancatelli@schema31.it> Cc: "freebsd-virtualization@freebsd.org" <freebsd-virtualization@freebsd.org> Subject: Re: bhyve - ESXi comparison part 2 Message-ID: <52EC6EA0.3000200@freebsd.org> In-Reply-To: <CADfWLekdn3Qz=wC2XkdP8ugd44P5scz-3EXamc5rs9psHfcohQ@mail.gmail.com> References: <CADfWLekdn3Qz=wC2XkdP8ugd44P5scz-3EXamc5rs9psHfcohQ@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Andrea, > Here comes the part 2 of our bhyve - ESXi comparison. Excellent work :) My take is that there is more general hypervisor overhead with bhyve. Given that both user and system times from the benchmark are almost uniformally larger for bhyve in all tests points to this. There has been work in CURRENT to allow the host to have a much lower clock, which could reduce hopefully a large piece of that overhead. The 20 x single CPU benchmark is really an ESXi vs FreeBSD filesystem test - the user/system times show the same penalty as in the first benchmark. The final benchmark points to how effective the ESXi scheduler is under heavy load and with multiprocessor guests. I suspect it goes to great lengths to avoid the 'lockholder preemption problem' - this is pointed to by the fact that the -P option with bhvye allows it to now complete the test, along with the large amount of time accounted as system time. later, Peter.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?52EC6EA0.3000200>