Whenever I read Red Hat material they seem obsessed with OpenStack, but I don’t actually know of anyone using it. Most of my professional contacts are either SF hipsters using hipster tech (e.g. Heroku, Kubernetes), or work at large companies that run their own infrastructure (e.g. Google, Amazon). So I don’t exactly have a representative sample to judge OpenStack’s mass market adoption.
I am a software engineer working for Red Hat on OpenStack, both on the upstream codebase and on Red Hat’s downstream Red Hat OpenStack Platform product. I must clarify that I announce this for disclosure, I am not a member of the press team.
Red Hat OpenStack Platform is certainly widely adopted by plenty of customers ranging from large scale compute users to telcos who like the NFV side of things and smaller scale shops who just want their own private cloud.
I am not at liberty to disclose many customer names, for that I’ll refer you to our PR department, but here’s something I found that was in our public success stories section of our website. That being said, OpenStack is a popular platform with a very wide install base across industries.
For high level information about Red Hat OpenStack Platform, see here.
For the upstream organization, here.
We have tried to look into it a few years back… It was not good. Basically an overengineered bigco consultant’s dream. Rackspace supposedly used it a lot, but not Dreamhost or DigitalOcean. It tries to play catch up with Amazon for some reason.
Has anything changed?
I don’t know about WW experience, but I know several russian companies with big clouds on OpenStack. We have a some experience with that painful solution. (Rabbitmq, for example, is thin place on big clouds (> 500 compute nodes in one cell)) Guys from OpenStack reinvent VRRP itself and when rabbitmq connection both network node and ampq server is broken - we get situation, when two routers with VRRP in MASTER state. Solutions from community - bump your erlang and rabbitmq version to upsteram and try. And many, many interesting things.
I worked for a hosting company that use OpenStack (https://www.openstack.org/marketplace/public-clouds/). Long story short, it’s used by a lot of companies, I have been to a couple of OpenStack Summit conferences and they are thousands of people there.
However, it’s not simple, very complex to maintain, hard to troubleshoot, upgrading to a new version is a pain in the ass, the number of projects under the OpenStack foundation could be overwhelming.
I worked at two companies with OpenStack. One was in New Zealand and they basically hemorrhaged money trying to start up a cloud provider using OpenStack. It was god awful. Machines constantly fell over, CPUs would get pegged for no reason, the built in permission model gave an admin token to almost everything (like basic monitoring).
My manager was really good and he had accounts on Redhat’s, Cannonical’s, Rackspace’s and others OpenStack implementation. We found tons of custom proprietary plugins that weren’t open sourced (that we could tell). There was no CVE mailing list; I had to write a script to parse the launchpad bugs so we could get CVEs. The debian repo/debs were more up to date than Ubuntu’s. Even Debian’s took a few weeks to patch CVEs, so we talked about creating our own build system. Standing up a dev environment was next to impossible because there were too many VMs to run on one desktop (even when using LXE containers instead of VirtualBox) and also open-vswitch can go die in a fire. Updates to parts of the software involved putting your patch on a wiki page (no git merge request system for many components).
After that job I worked for a Seattle company that tried to migrate off AWS onto OpenStack. It was a fucking disaster. They couldn’t keep that cluster stable worth shit and eventually dropped it and went to DC/OS. DC/OS might be a resource hog, but it was stable and it worked for scheduling containers. They replaced the entire OpenStack cluster with DC/OS nodes and it turned into a company wide platform for everything.
Fuck OpenStack. Burn it with a god damn fire. It’s a piece of shit that you should use never.
RackSpace still uses it, and they’re a relatively large service provider similar to AWS (in their offerings, not userbase).
Rackspace is now also an AWS/Azure/GoogleCloud consulting company. They seem to have given up on the direct competition part.
I’m using RedHat’s other virtualization product, oVirt, for my personal homelab. It’s quite smooth and very well made.
I looked into OpenStack as well, but there’s so many disparate components it cases the installation instructions to be way too complex. I’ve been turned off by the sheer complexity of all the components. (Many of which are optional, so that’s even extra complexity.)
I plan on deploying OpenShift soon too, to learn me some Kubernetes. :)
I use it (3-node cluster) in my homelab. I’ve deployed it at a handful of customer sites, all < 50 nodes, as an alternative to VMWare, and everyone seems happy with it.
In academia, the one big deployment I know of is a 99-node cluster at MIT.
CERN (http://superuser.openstack.org/articles/cern-expanding-cloud-universe/) and the NSA (https://blog.rackspace.com/nsa-opens-the-kimono-on-its-openstack) use it. Quite large deployments too.
I’ve managed teams deploying it in large scale for internal clouds as well.
It is a PITA if you use all the bells and whistles. My advice would be keep it simple. VMs still have lots of value… Containers haven’t hit the enterprise quite yet.
We use OpenStack for all our internal virtual machines (which are quite a lot). I think it’s quite (too?) complex to setup, but the ease of use after the initial setup is really good. There are a lot of OpenStack-compatible images available and it also isn’t difficult to make them yourself for some BSD’s of lesser used Linux distro’s. I really like the automatic SSH-key deployment, IP address configuration (we use public IPv4 and IPv6) and ACL/firewall in the gui. I know some more organizations that use OpenStack, but most (like us) aren’t very public about it. We actually added more/new nodes to it recently :). Hope this helps!
I remember taking a look at it when it first came out (I worked at Brightbox at the time, whilst we were laying the foundation for the cloud product) and thinking it looked like something designed by committee rather than something solving discrete problems properly. Not seen anything since to change my opinion of it.
I used OpenStack as the basis for a IaaS-generator solution, but it was terrible. nova.conf had 800+ options but there were really only about 15 useful configurations, so most of the work was setting those options to match the rest of the other options. I started the project over again with OpenNebula and got MUCH better results. It has been in production for over 3years now running non-stop customer workloads.
I attempted to switch our services from Amazon s3 to the OpenStack version but found it very confusing. Ended up giving up because I couldn’t find the info I needed.
Alta3 Research does. It has been a workhorse for deploying training environments (I do side work for them).
They have a nice video on how openstack worked well for them and cut costs.
A couple of German universities use it to give researchers easy access to cloud computing: https://www.bw-cloud.org/en/project
We tried it at $WORK, this was 3 years ago or so. Super hassle to stand up. You couldn’t upgrade it, you had to destroy your deployment and redeploy. We bought a product that wrapped around OpenStack to help manage configurations and upgrades (I forget the name, they were bought by Cisco) until they went away.
I tried using it just for S3 compatible storage. Got it set up and working, shut it down, and I wouldn’t start back up again. I gave up at that point.