* *Size*: VMs are very large which makes them impractical to store and transfer.
* *Performance*: running VMs consumes significant CPU and memory, which makes them impractical in many scenarios, for example local development of multi-tier applications, and
* *Portability*: competing VM environments don't play well with each other. Although conversion tools do exist, they are limited and add even more overhead.
* *Hardware-centric*: VMs were designed with machine operators in mind, not software developers. As a result, they offer very limited tooling for what developers need most:
building, testing and running their software. For example, VMs offer no facilities for application versioning, monitoring, configuration, logging or service discovery.
By contrast, Docker relies on a different sandboxing method known as *containerization*. Unlike traditional virtualization,
containerization takes place at the kernel level. Most modern operating system kernels now support the primitives necessary
for containerization, including Linux with [openvz](http://openvz.org), [vserver](http://linux-vserver.org) and more recently [lxc](http://lxc.sourceforge.net),
Solaris with [zones](http://docs.oracle.com/cd/E26502_01/html/E29024/preface-1.html#scrolltoc) and FreeBSD with [Jails](http://www.freebsd.org/doc/handbook/jails.html).
A common problem for developers is the difficulty of managing all their application's dependencies in a simple and automated way.
This is usually difficult for several reasons:
* *Cross-platform dependencies*. Modern applications often depend on a combination of system libraries and binaries, language-specific packages, framework-specific modules,
internal components developed for another project, etc. These dependencies live in different "worlds" and require different tools - these tools typically don't work
well with each other, requiring awkward custom integrations.
* Conflicting dependencies. Different applications may depend on different versions of the same dependency. Packaging tools handle these situations with various degrees of ease -
* Custom dependencies. A developer may need to prepare a custom version of their application's dependency. Some packaging systems can handle custom versions of a dependency,
others can't - and all of them handle it differently.
Docker defines a build as running a sequence of unix commands, one after the other, in the same container. Build commands modify the contents of the container
(usually by installing new files on the filesystem), the next command modifies it some more, etc. Since each build command inherits the result of the previous
commands, the *order* in which the commands are executed expresses *dependencies*.
Under the hood, Docker is built on the following components:
* The [cgroup](http://blog.dotcloud.com/kernel-secrets-from-the-paas-garage-part-24-c) and [namespacing](http://blog.dotcloud.com/under-the-hood-linux-kernels-on-dotcloud-part) capabilities of the Linux kernel;
* [AUFS](http://aufs.sourceforge.net/aufs.html), a powerful union filesystem with copy-on-write capabilities;
* The [Go](http://golang.org) programming language;
* [lxc](http://lxc.sourceforge.net/), a set of convenience scripts to simplify the creation of linux containers.
Docker defines a unit of software delivery called a Standard Container. The goal of a Standard Container is to encapsulate a software component and all its dependencies in
a format that is self-describing and portable, so that any compliant runtime can run it without extra dependencies, regardless of the underlying machine and the contents of the container.
The spec for Standard Containers is currently a work in progress, but it is very straightforward. It mostly defines 1) an image format, 2) a set of standard operations, and 3) an execution environment.
A great analogy for this is the shipping container. Just like how Standard Containers are a fundamental unit of software delivery, shipping containers (http://bricks.argz.com/ins/7823-1/12) are a fundamental unit of physical delivery.
Just like shipping containers, Standard Containers define a set of STANDARD OPERATIONS. Shipping containers can be lifted, stacked, locked, loaded, unloaded and labelled. Similarly, standard containers can be started, stopped, copied, snapshotted, downloaded, uploaded and tagged.
Just like shipping containers, Standard Containers are CONTENT-AGNOSTIC: all standard operations have the same effect regardless of the contents. A shipping container will be stacked in exactly the same way whether it contains Vietnamese powder coffee or spare Maserati parts. Similarly, Standard Containers are started or uploaded in the same way whether they contain a postgres database, a php application with its dependencies and application server, or Java build artifacts.
Both types of containers are INFRASTRUCTURE-AGNOSTIC: they can be transported to thousands of facilities around the world, and manipulated by a wide variety of equipment. A shipping container can be packed in a factory in Ukraine, transported by truck to the nearest routing center, stacked onto a train, loaded into a German boat by an Australian-built crane, stored in a warehouse at a US facility, etc. Similarly, a standard container can be bundled on my laptop, uploaded to S3, downloaded, run and snapshotted by a build server at Equinix in Virginia, uploaded to 10 staging servers in a home-made Openstack cluster, then sent to 30 production instances across 3 EC2 regions.
### 4. DESIGNED FOR AUTOMATION
Because they offer the same standard operations regardless of content and infrastructure, Standard Containers, just like their physical counterpart, are extremely well-suited for automation. In fact, you could say automation is their secret weapon.
Many things that once required time-consuming and error-prone human effort can now be programmed. Before shipping containers, a bag of powder coffee was hauled, dragged, dropped, rolled and stacked by 10 different people in 10 different locations by the time it reached its destination. 1 out of 50 disappeared. 1 out of 20 was damaged. The process was slow, inefficient and cost a fortune - and was entirely different depending on the facility and the type of goods.
Similarly, before Standard Containers, by the time a software component ran in production, it had been individually built, configured, bundled, documented, patched, vendored, templated, tweaked and instrumented by 10 different people on 10 different computers. Builds failed, libraries conflicted, mirrors crashed, post-it notes were lost, logs were misplaced, cluster updates were half-broken. The process was slow, inefficient and cost a fortune - and was entirely different depending on the language and infrastructure provider.
There are 17 million shipping containers in existence, packed with every physical good imaginable. Every single one of them can be loaded onto the same boats, by the same cranes, in the same facilities, and sent anywhere in the World with incredible efficiency. It is embarrassing to think that a 30 ton shipment of coffee can safely travel half-way across the World in *less time* than it takes a software team to deliver its code from one datacenter to another sitting 10 miles away.
