Why Virtualization Fragmentation Sucks
Mass adoption of virtualization is upon us. A plethora of virtualization vendors have entered the market. Each has a slightly different set of features, disk formats, configuration files, and guest kernel drivers. As concepts such as Virtual Appliances become mainstream, software vendors are faced with new challenges. Previously, software vendors had to port their application to multiple operating systems: Solaris, Linux, AIX, etc. The new "port" becomes one where software vendors will be expected to produce images that drop-in to VMware, Xen, Parallels, SLES, RHEL, and even Microsoft Virtual Server.
This paper will explore the state of existing virtualization technology in meeting the goal of providing ready-to-run guest images. This includes: comparing, contrasting, and poking fun at virtual disk formats; bemoaning the assortment of kernel drivers needed to improve performance in a guest (vmware-tools, paravirt drivers...); dark muttering about incompatibilities between Xen guests and hosts; and lamenting all the different configuration files that define a guest.