All in the name of Liberty: OpenStack 12 set free! The 12th version of OpenStack, Liberty, is on the streets. The OpenStack foundation reckons it's kicked some big goals this time around, especially in terms of security thanks to refined role-based controls and scalability thanks to a hotter Nova. Plenty of OpenStack users are service providers, so it makes sense that the new …
It's piss awful.
However that's not necessarily down to the technology, but the implementation - however if you can make it that shite then there's something wrong with the tech too IMHO.
We're an MSP that's focuses mainly on MS tech. Our upstream network partner (we resell their products as a boutique ISP) that we have colo with offered us the opportunity to be the first client on their new OpenStack environment.
Hours spent getting drivers for Windows 2008 R2 and injecting them in. Had the provider reboot our instance a few times by mistake. Advised there's no firewall between the instance and our colo other than our main colo firewall. Total, utter bollocks. Again, hours spent trying to fix an issue after their accidentially reset the config on this non-existant firewall so turning it one-way (that passed ICMP both ways just to really throw me off!) which totally screwed up our Exchange DAG that spanned into the OpenStack instance.
And the performance... my $God. We've actually given up trying to backup from the OpenStack instance now entirely. The read performance is terrible. VSS backup of Exchange on the VM on our HyperV colo - < hour. OpenStack, gets to >5 hours and then times out. Apparently they have a "Ceph" cluster with SSD's and excellent IOPS - but I'm clearly not seeing it.
Sorry, but I've rolled my own vSphere (3.5 - 5.5) and HyperV (2008R2 - 2012 R2) as well as used Azure and vCloud Air and OpenStack, in the way our provider configured it is just a dead duck. Nothing good at all came from us using it.
We've since brought a new bit of tin and are in the process of rebuilding the instances we have on OpenStack so we can decommission.
Appreciate that this is probably down to how it's been implemented - but these guys aren't spanners. They've FOSS nuts who develop their own Linux based firewall and run as an ISP. All Linux nuts.
Sure, they've probably messed things up - but I can't remember a time when I deployed VMWare of Microsoft products for the first time and things ended up being so... well shit.
Sure if you're a massive company with a lot of FOSS resource you can dedicate you can probably craft your own cloudy stack and OpenStack is the best way of doing this.
Everyone else, seriously, I'd recommend paying HP to use Helion or a similar provider, or just use Microsoft's Azure Stack onpremise or use VMWare's vSphere stack instead.
After the rest of the Infrastructure team here started calling it OpenShat I think I decided we weren't going to look into it any futher...
most of the config is obscure flat text files...which can be version controlled like code files, centrally managed, duplicated from dev/test environments to live, tagged and released to live in an orderly and therefore safe fashion after testing, parsed by standard tools, munged with other standard tools, easy to back up and easy to see diffs between backups if you don't source-control/tag/release, and, apart from having to learn what needs to go in them by using the "man" pages for them, allow for a full, flexible and scalable experience, as one would expect from a POSIX/unix application.
FTFY :-)
(N.B. - I haven't actually used openstack, it may well be an absolute pain to configure...I simply take exception to the idea of "plain text configuration files" being inherently a bad thing)
Couldn't agree with you more. I actually prefer text files over anything else. At the very least I can just pop them open in vi and not be overwhelmed by hundreds of lines of cruft surrounding 10-15 actual lines of config. My policy is that if a config takes up more than a megabyte fo disk, you might want to re-write it and/or split it into smaller pieces.
While we haven't run anything in production on OpenStack, we are actively exploring OpenStack as an option to replace our extremely expensive VMWare environment. We have an OpenStack pilot running on an Intel NUC and the performance is quite good. It was dead simple to setup using the instructions found here: https://www.rdoproject.org/Quickstart
I'm very optimistic on OpenStack's roadmap. Redhat says Liberty will be the first release to get longer term support, which makes us more comfortable when or if it comes time to place the bet. Quite frankly, I desperately want to see OpenStack as a viable IaaS. I think just about everything else (sans our Windows print server) in our environment can be docker'ized to run in OpenShfit version v3, which of course would run on top of OpenStack. Scalable micro services, run through an automated infrastructure tool PaaS like OpenShift seems to be the future.
Bless them, every time they come across OpenStack they completely miss the point. When they use the term "not enterprise" they forget that it is not aimed at workloads running on "enterprise" platforms (i.e, VMware) but at the newer 12 factor style apps, they are comparing apples to oranges. They also don't seem to understand the bigger picture where if OpenStack's aspirations come off the entire relationship between customer and IT vendors/IT department is flipped on its head.
They also conveniently forget that most customers wont take trunk OpenStack but use a distribution (in the same way very few people roll their own Linux distribution) where the vendor has solved many of the challenges they claim OpenStack has.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
