Solaris Containers (including Solaris Zones) is an implementation of operating system-level virtualization technology for x86 and SPARC systems, first released publicly in February 2004 in build 51 beta of Solaris 10, and subsequently in the first full release of Solaris 10, 2005. It is present in newer OpenSolaris based distributions, like OpenIndiana and Solaris 11 Express.
A Solaris Container is the combination of system resource controls and the boundary separation provided by zones. Zones act as completely isolated virtual servers within a single operating system instance. By consolidating multiple sets of application services onto one system and by placing each into isolated virtual server containers, system administrators can reduce cost and provide most of the same protections of separate machines on a single machine.
Contents |
There is always one zone defined, named the "global" zone. Zones hosted by a global zone are known as "non-global zones" but are sometimes just called "zones". The term "local zone" is specifically discouraged, since in this usage "local" is not an antonym of "global". The global zone encompasses all processes running on the system, whether or not these processes are running within a non-global zone. Unless otherwise noted, "zone" will refer to non-global zones in this article.
Each zone has its own node name, virtual network interfaces, and storage assigned to it; there is no requirement for a zone to have any minimum amount of dedicated hardware other than the disk storage necessary for its unique configuration. Specifically, it does not require a dedicated CPU, memory, physical network interface or HBA, although any of these can be allocated specifically to one zone.
Each zone has a security boundary surrounding it which prevents a process associated with one zone from interacting with or observing processes in other zones. Each zone can be configured with its own separate user list. The system automatically manages user ID conflicts; that is, two zones on a system could have a user ID 10000 defined, and each would be mapped to its own unique global identifier.
A zone can be assigned to a resource pool (processor set plus scheduling class) to guarantee certain usage, or can be given shares via fair-share scheduling. A zone can be in one of the following states:
Some programs cannot be executed from within a non-global zone; typically this is because the application requires privileges that cannot be granted within a container. As a zone does not have its own separate kernel (in contrast to a hardware virtual machine), applications that require direct manipulation of kernel features, such as the ability to directly read or alter kernel memory space, may not work inside of a container.
Zones induce a very low overhead on CPU and memory. Currently a maximum of 8191 non-global zones can be created within a single operating system instance. "Sparse Zones", in which most filesystem content is shared with the global zone, can take as little as 50MB of disk space. "Whole Root Zones", in which each zone has its own copy of its operating system files, may occupy anywhere from several hundred megabytes to several gigabytes, depending on installed software.
Even with Whole Root Zones, disk space requirements can be negligible if the zone's OS file system is a ZFS clone of the global zone image, since only the blocks different from a snapshot image need to be stored on disk; this method also makes it possible to create new zones in a few seconds.
Although all zones on the system share a common kernel, an additional feature set has been added called branded zones (BrandZ for short). This allows individual zones to behave in a manner other than the default brand of the global zone. The existing brands (October 2009) can be grouped into two categories:
The brand for a zone is set at the time the zone is created. The second category is implemented with interposition points within the OS kernel that can be used to change the behavior of syscalls, process loading, thread creation, and other elements.
For the 'lx' brand, libraries from Red Hat 3 or an equivalent distribution such as CentOS are required to complete the emulated environment.
The Solaris operating system provides man pages for Solaris Containers by default; more detailed documentation can be found at various on-line technical resources.
The first published document and hands-on reference for Solaris Zones was written in February 2004 by Dennis Clarke at Blastwave.org, providing the essentials to getting started.[2] This document was greatly expanded upon by Brendan Gregg in July 2005.[3] The Solaris 8 and Solaris 9 Containers were documented in detail by Dennis Clarke at Blastwave(tm) again in April 2008 and this has become a simple How To style guide that can get people started with Solaris Containers in a production setting.[4] The Blastwave Solaris 8 and Solaris 9 Containers document was very early in the release cycle of the Solaris Containers technology and the actions and implementation at Blastwave resulted in a followup by Sun Microsystems marketing.[5] More extensive documentation may be found at the Sun Microsystems documentation site,[6] the Sun BluePrints Archive,[7] and the Sun Solaris Containers Learning Center.[8]
The standard Solaris NFS server is implemented in the kernel, and cannot be used for exports within non-global zones.[9][10] Third party NFS server software that is not implemented in the Solaris kernel may work.
Branded zones are not supported on the sun4us architecture (Fujitsu PRIMEPOWER servers).[11]
|