Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread Michael Rasmussen
On Mon, 13 Apr 2015 18:48:44 +0200 (CEST) Tobias Oetiker t...@oetiker.ch wrote: BUT what worries me, is that between omnios r10 and r14 (r12 actually) there is a factor of 1000 or more !!! Just a long shot. Has the compiler and linker used to build kernel and modules changed between r10 -

Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread Dan McDonald
On Apr 13, 2015, at 1:28 PM, Michael Rasmussen m...@miras.org wrote: On Mon, 13 Apr 2015 18:48:44 +0200 (CEST) Tobias Oetiker t...@oetiker.ch wrote: BUT what worries me, is that between omnios r10 and r14 (r12 actually) there is a factor of 1000 or more !!! Just a long shot. Has the

Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread Michael Rasmussen
On Mon, 13 Apr 2015 16:25:42 + John Barfield john.barfi...@bissinc.com wrote: I’ll try to pull some numbers but the truth is that debian based guest versions perform WAY better than CentOS guests. Are the kernels identical in Debian and CentOS? -- Hilsen/Regards Michael Rasmussen

Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread John Barfield
CentOS still uses the traditional 2.6 numbering convention and back ports modules and features that they feel should be back ported while Ubuntu/Debian is following the kernel.org release cycle. (For the most part). In other words…its hard to tell the differences. I spent the better part of

Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread Tobias Oetiker
Michael, Today Michael Rasmussen wrote: On Mon, 13 Apr 2015 16:25:42 + John Barfield john.barfi...@bissinc.com wrote: I?ll try to pull some numbers but the truth is that debian based guest versions perform WAY better than CentOS guests. Are the kernels identical in Debian and

Re: [OmniOS-discuss] [smartos-discuss] kvm network performance survey

2015-04-13 Thread John Barfield
I can confirm that I’ve seen the same behaviors. I’ll try to pull some numbers but the truth is that debian based guest versions perform WAY better than CentOS guests. From all of my testing IMHO it is the host (illumos) virtio implementation. I only say this because I see such drastic