Comments on: A Virtual OS for My Hypervisor Please http://www.linux-mag.com/id/7523/ Open Source, Open Standards Sat, 05 Oct 2013 13:48:18 +0000 hourly 1 http://wordpress.org/?v=3.1 By: perfect world international deutsch http://www.linux-mag.com/id/7523/#comment-63547 perfect world international deutsch Wed, 30 Nov 2011 22:51:46 +0000 http://www.linux-mag.com/id/7523/#comment-63547 Hello there, simply turned into aware of your blog through Google, and found that it's really informative. I'm gonna be careful for brussels. I will appreciate should you proceed this in future. Many other folks will probably be benefited out of your writing. Cheers! Hello there, simply turned into aware of your blog through Google, and found that it’s really informative. I’m gonna be careful for brussels. I will appreciate should you proceed this in future. Many other folks will probably be benefited out of your writing. Cheers!

]]> By: unclesmrgol http://www.linux-mag.com/id/7523/#comment-6997 unclesmrgol Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7523/#comment-6997 <p>If you\'ve examined VMWare\'s ESXi/vSphere offering closely, you will discover that they have built a Virtual OS for their hypervisor, and that virtual OS is based on the Linux kernel. </p> <p>They\'ve added device drivers for their proprietary filesystem, a bunch of tools to manipulate virtual machines, but underneath is a stock kernel. If anyone is concerned about benchmarks, it\'s got to be VMWare, what with Microsoft breathing down their backs. They are big enough that if they thought the Linux kernel wasn\'t cutting it, they\'d go build one of their own. But they haven\'t, so the kernel crew has to be doing something good which you don\'t seem to have noticed. </p> If you\’ve examined VMWare\’s ESXi/vSphere offering closely, you will discover that they have built a Virtual OS for their hypervisor, and that virtual OS is based on the Linux kernel.

They\’ve added device drivers for their proprietary filesystem, a bunch of tools to manipulate virtual machines, but underneath is a stock kernel. If anyone is concerned about benchmarks, it\’s got to be VMWare, what with Microsoft breathing down their backs. They are big enough that if they thought the Linux kernel wasn\’t cutting it, they\’d go build one of their own. But they haven\’t, so the kernel crew has to be doing something good which you don\’t seem to have noticed.

]]> By: zish http://www.linux-mag.com/id/7523/#comment-6998 zish Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7523/#comment-6998 <p>This is the beauty behind Xen and the Linux kernel, IMHO. The hypervisor itself employs something called the XenBus, which guest OS\'es can use for Network and Block I/O.<br /> If your guest OS employs a \"Xen-aware\" kernel and drivers, the performance is at or near \"native\". The modern Linux and Solaris kernels have this capability. I believe that *BSD has it as well, but I have not bothered to check. Novell and Citrix have also released drivers for Windows that support access to the Xen bus for Block and Network I/O. I\'ve used the Novell drivers quite extensively in production environments, with great success.<br /> Of course, you don\'t -need- to make use of the XenBus, since Xen employs Qemu and the Bochs BIOS to provide emulated PCI devices. But you definitely get a huge performance increase if you do, and the OS is then fully \"Hypervisor-aware\".</p> <p>Other than the above, I don\'t think any additional optimizations should be needed. IMHO, I\'d still like to be able to manage my \"virtual\" machines in the same manner as their \"physical\" counterparts. </p> This is the beauty behind Xen and the Linux kernel, IMHO. The hypervisor itself employs something called the XenBus, which guest OS\’es can use for Network and Block I/O.
If your guest OS employs a \”Xen-aware\” kernel and drivers, the performance is at or near \”native\”. The modern Linux and Solaris kernels have this capability. I believe that *BSD has it as well, but I have not bothered to check. Novell and Citrix have also released drivers for Windows that support access to the Xen bus for Block and Network I/O. I\’ve used the Novell drivers quite extensively in production environments, with great success.
Of course, you don\’t -need- to make use of the XenBus, since Xen employs Qemu and the Bochs BIOS to provide emulated PCI devices. But you definitely get a huge performance increase if you do, and the OS is then fully \”Hypervisor-aware\”.

Other than the above, I don\’t think any additional optimizations should be needed. IMHO, I\’d still like to be able to manage my \”virtual\” machines in the same manner as their \”physical\” counterparts.

]]> By: cjcox http://www.linux-mag.com/id/7523/#comment-6999 cjcox Wed, 30 Nov -0001 00:00:00 +0000 http://www.linux-mag.com/id/7523/#comment-6999 <p>On the enterprise side Novell\'s SLES has had a VMware VMI enabled kernel, BUT it was NOT part of the GA for SLES 10, so it\'s an add-on and only for 32bit.</p> <p>AFAIK, the VMware VMI stuff is part of newer kernels... however my own experience has shown it to be buggy at best. Ideally though, you just install a fairly new Linux distro and turn on paravirtualiation on the ESX option panel for the guest.</p> <p>The common workaround for most of the bugs it to change the guest to use just one virtual cpu... sort of limiting, but keeps things stable longer. However, I find that even so, eventually, it will lock up hard.</p> <p>My SLES10 SP2 guests with the vmi 32bit kernel work just fine. But some have noted some problems with massively multi-threaded java apps when the guest has multiple CPUs on SLES10 SP2... and again, the solution is to reduce the number of virtual CPUs to one.</p> <p>Certainly more work needed.. and obviously VMI encourages VMware lock-in. </p> On the enterprise side Novell\’s SLES has had a VMware VMI enabled kernel, BUT it was NOT part of the GA for SLES 10, so it\’s an add-on and only for 32bit.

AFAIK, the VMware VMI stuff is part of newer kernels… however my own experience has shown it to be buggy at best. Ideally though, you just install a fairly new Linux distro and turn on paravirtualiation on the ESX option panel for the guest.

The common workaround for most of the bugs it to change the guest to use just one virtual cpu… sort of limiting, but keeps things stable longer. However, I find that even so, eventually, it will lock up hard.

My SLES10 SP2 guests with the vmi 32bit kernel work just fine. But some have noted some problems with massively multi-threaded java apps when the guest has multiple CPUs on SLES10 SP2… and again, the solution is to reduce the number of virtual CPUs to one.

Certainly more work needed.. and obviously VMI encourages VMware lock-in.

]]>