A Brief History of Containers

With a Few Noteworthy Tangents

About Me

  • Nathan Murthy
     
  • Software Engineer at Tesla / SolarCity
     
  • Interests: distributed systems, microservices, big data, machine learning, clean energy for the masses

"You are always in a container. Even
  when you don't run containers, you
  are in a container."

 

                                                  (Sipe 2016)

docker = container

docker = automation

docker ≠ automation

https://www.pinterest.com/email2sekhar/devops-understanding/

docker  ≠

1. OS-Level Virtualization
 

2. Hardware Virtualization

Two Historical Trends in Computing

History of OS-Level Virtualization

1979-1982 UNIX chroot
1998 FreeBSD Jail
2001 Linux VServer
2002 Linux namespaces
2005 Sun Microsystems: Solaris Containers / Solaris Zones
2005 OpenVZ
2007 Google: cgroups
2007 PaaS: Heroku, Joyent, CloudFoundry
2008 LXC
2010 systemd
2013 Docker
2015 CoreOS: Application Container Spec (appC, rkt)
2015 Open Container Initiative: lmctfy, libcontainer, runC

History of Hardware Virtualization

1966-1972 IBM (CP/CMS)
1974 Popek & Goldberg coin the term "hypervisor" in
Formal Requirements for Virtualizable Third Generation Architectures
1989 Insignia SoftPC
1995 Sun Microsystems: Java Virtual Machine
1998 VMWare
2003 Xen Hypervisor
2006 Amazon EC2
2007 Sun: Logical Domains
(Oracle VM Server for SPARC and x86)
2007 Linux KVM
2008 Microsoft Hyper-V

chroot

$ man chroot

jail

jail

A Jail in FreeBSD-speak is one or more tasks with the same kernel Jail-ID, bound on zero or more IP addresses, having the same chroot-environment.

https://erdgeist.org/arts/software/ezjail/

$ man jail

Linux-VServer

Linux Namespaces

Namespace Isolates
Cgroup* Cgroup root directory
IPC System V IPC, POSIX message queues
Network Network devices, stacks, ports, etc.
Mount Mount points
PID Process IDs
User User and group IDs
UTS Hostname and NIS domain name

Linux provides the following namespaces:

http://man7.org/linux/man-pages/man7/namespaces.7.html

$ sleep 20 &
[1] 2325

$ sudo tree /proc/2325/ns
/proc/2325/ns
├── ipc -> ipc:[4026531839]
├── mnt -> mnt:[4026531840]
├── net -> net:[4026531956]
├── pid -> pid:[4026531836]
├── user -> user:[4026531837]
└── uts -> uts:[4026531838]

Sun introduces Solaris Containers and Solaris Zones, an alternative to their virtual machine technology.

Solaris Containers

"chroot on steroids"

(Schmidt 2006)

OpenVZ

cgroups

  • Contribution by Google to Linux kernel
     
  • Originally called "process containers"
     
  • Renamed to  "cgroups"
     
  • Merged into kernel 2.6.24

cgroups

$ ls -l /sys/fs/cgroup

drwxr-xr-x 4 root root 0 Feb 18 21:55 blkio
drwxr-xr-x 4 root root 0 Feb 18 21:55 cpu
drwxr-xr-x 4 root root 0 Feb 18 21:55 cpuacct
drwxr-xr-x 4 root root 0 Feb 18 21:55 cpuset
drwxr-xr-x 4 root root 0 Feb 18 21:55 devices
drwxr-xr-x 5 root root 0 Feb 18 21:55 freezer
drwxr-xr-x 4 root root 0 Feb 18 21:55 hugetlb
drwxr-xr-x 4 root root 0 Feb 18 21:55 memory
drwxr-xr-x 4 root root 0 Feb 18 21:55 perf_event
drwxr-xr-x 4 root root 0 Feb 18 21:55 systemd

namespaces + cgroups

cgroup: limits how much you can use
 

namespace: limits how much you can

see (and therefore use)

 

(Sipe 2016)

LXC

(Hykes 2013)

Newer kids on the block

???

Landscape Today

History of OS-Level Virtualization

1979-1982 UNIX chroot
1998 FreeBSD Jail
2001 Linux VServer
2002 Linux namespaces
2005 Sun Microsystems: Solaris Containers / Solaris Zones
2005 OpenVZ
2007 Google: cgroups
2007 PaaS: Heroku, Joyent, CloudFoundry
2008 LXC
2010 systemd
2013 Docker
2015 CoreOS: Application Container Spec (appC, rkt)
2015 Open Container Initiative: lmctfy, libcontainer, runC

Containers are all about

  • OS-Level Virtualization
     
  • Visibility
     
  • Isolation
     
  • Resource Management
  • Automation (?)  
  • Microservices  

Hmm, but you may be just as well off with tools like wget, zip, tar, upstart, systemd, and the gamut of commands already available to you on your *nix host machine. 

Depends on scale and granularity.

Containers at Scale

 

"One of Borg’s primary goals is to make efficient use of Google’s fleet of machines, which represents a significant financial investment: increasing utilization by a few percentage points can save millions of dollars.
...
Borg users request CPU in units of milli-cores, and memory and disk space in bytes."


 

 

 

Verma et al. "Large-scale cluster management at Google with Borg" in ACM EuroSys (2015).

Inspired by:

 

1. Rani Osnat: http://blog.aquasec.com/a-brief-history-of-containers-from-1970s-chroot-to-docker-2016

 

2. Bryan Cantrill: http://www.slideshare.net/bcantrill/run-containers-on-bare-metal-already

 

Q+A

  • @natemurthy
     
  • namurthy@tesla.com
     
  • github.com/natemurthy