CoreOS is a slimmed-down Linux distribution designed for easy creation of lots of OS instances. We like the concept.
CoreOS uses Docker to deploy applications in virtual containers; it also features a management communications bus, and group instance management.
Rackspace, Amazon Web Services (AWS), GoogleComputeEngine (GCE), and Brightbox are early cloud compute providers compatible with CoreOS and with specific deployment capacity for CoreOS. We tried Rackspace and AWS, and also some local "fleet" deployments.
+ ALSO ON NETWORK WORLD: Red Hat Enterprise Linux 7 anchors enterprise focused ecosystem +
CoreOS is skinny. We questioned its claims of less overall memory used, and wondered if it was stripped to the point of uselessness. We found that, yes, it saves a critical amount of memory (for some), and no, it's tremendously Spartan, but pretty useful in certain situations.
CoreOS has many similarities with Ubuntu. They're both free and GPLv3 licensed. Ubuntu 14.04 and CoreOS share the same kernel. Both are easily customizable, and no doubt you can make your own version. But CoreOS shuns about half of the processes that Ubuntu attaches by default.
If you're a critic of the bloatware inside many operating systems instances, CoreOS might be for you. In testing, we found it highly efficient. It's all Linux kernel-all-the-time, and if your organization is OS-savvy, you might like what you see in terms of performance and scale.
Security could be an issue
CoreOS uses curl for communications and SSL, and we recommend adding a standard, best-practices external SSL certificate authority for instance orchestration. Otherwise, you'll be madly generating and managing SSL relationships among a dynamic number of instances. CoreOS sends updates using signed certificates, too.
With this added SSL security control, your ability to scale efficiently is but a few scripts away. Here's the place where your investment in SSL certs and chains of authority back to a root cert is a good idea. It adds to the overhead, of course, to use SSL for what might otherwise be considered "trivial" instances. All the bits needed for rapid secure communications with SSL are there, and documented, and wagged in your face. Do it.
What You Get
CoreOS is a stripped-down Linux distro designed for rapidly deployed Spartan instance use. The concept is to have a distro that's bereft of the usual system memory and daemon leeches endemic to popular distributions and "ecosystems.'' This is especially true as popular distros get older and more "feature packed".
More available memory usually means more apps that can be run, and CoreOS is built to run them in containers. Along with its own communications bus--primitive as it is-- you get to run as many instances (and apps) as possible with the least amount of overhead and management drama.
For those leaning towards containerized instances, it blasts them in a controlled procedure, then monitors them for health. It's not tough to manage the life cycle of a CoreOS instance. RESTful commands do much of the heavy lifting.
Inside CoreOS is a Linux kernel, LXC capacity, and the etcd/etcd daemon service discovery/control daemon, along with Docker, the application containerization system, and systemd--the start/stop process controller that's replaced various initd (initial daemon) in many distros.
There is multiple instance management using fleet--a key benefit for those primarily starting pools and even oceans of instances of apps/OS instances on a regular basis.
Like Ubuntu and RedHat, it uses the systemd daemon as an interface control mechanism, and it's up to date with the same kernel used by Ubuntu 14.04 and RedHat EL7. Many of your updated systemd-based scripts will work without changes.
The fleetd is controlled by the user space command fleetctl and it instantiates processes, and the etcd daemon is a service discovery (like a communications bus) using etcdctl for monitoring--all at a low level and CLI-style.
The etcd is used to accept REST commands, using simple verbs. It uses a RESTful API set, and it's not Puppet, Chef, or other service bus communications bus controller, but a lean/tight communications methodology. It works and is understandable by Unix/Linux coders and admins.
A downside is that container and instance sprawl become amazingly easy. You can fire instances, huge number of them, at will. There aren't any clever system-wide monitoring mechanisms that will warn you that your accounting department will simply explode when they see your sprawl bill on AWS or GCE. Teardown isn't enforced--but it's not tough to do.
We did a test to determine the memory differences between Ubuntu 14.04 and CoreOS, configuring each OS as 1GB memory machines on the same platform. They reported the same kernel (Linux 3.12), and were used with default settings.
We found roughly 28% to 44% more memory available for apps with CoreOS -- before "swap" started churning the CPU/memory balances within the state machine.
This means an uptake in speed of execution for apps until they need I/O or other services, less memory churn and perhaps greater cache hits. Actual state machine performance improvements are dependent on how the app uses the host but we feel that the efficiencies of memory use and overall reduction in bloat (and security attack surface potential) are worth the drill.
These results were typical across AWS, GCE, and our own hosted platform that ran on a 60-core HP DL-580 Gen8. The HP server used could probably handle several hundred instances if we expanded the server's memory to its 6TB max--not counting Docker instances.
We could easily bring up a fleet of CoreOS instances, control it, feed it containers with unique IDs and IPs, make the containers do work (we did not exercise the containers), then shut them down, mostly with shell scripts rather than direct commands.
The suggested scripts serve as a template, and more templates are appearing, that allowed us to easily replicate functionality, so as to manage sprawl. If you're looking for instrumentation, get some glitzy UI elsewhere, and the same goes for high-vocabulary communications infrastructure.
Once you start adding daemons and widgetry, you're back to Ubuntu or RedHat.
And we warn that we could also make mistakes that were unrecoverable with equally high speed, and remind that there aren't any real safeguards except syntax checking, and the broad use of SSL keys.
You can make hundreds of OS instances, each with perhaps 100 Docker container apps all moving hopefully in a harmonious way. Crypt is used, which means you need your keys ready to submit to become su/root. Otherwise, you're on your own.
This is a skinny instance, bereft of frills and daemons-with-no-use. We found more memory and less potential for speed-slowing memory churn. Fewer widgets and daemonry also means a smaller attack surface. No bloat gives our engineer's instinctual desire to match resources with needs--no more and no less--more glee.
CoreOS largely means self-support, your own instrumentation, plentiful script building, and liberation from the pomposity and fatuousness of highly featured, general purpose, compute-engines.
Like to rough it? Want more real memory for actual instances? Don't mind a bit of heavy lifting? CoreOS might be for you.
Tom Henderson runs ExtremeLabs, Inc., in Bloomington, Ind. He can be reached at firstname.lastname@example.org.