Everything you know about OpenStack is wrong When someone talks about "the future of IT infrastructure delivery" the scope of topic is so broad that there's a really good chance they're working with an Etch-a-Sketch that lacks some of the knobs. IT isn't homogeneous. What works for enterprises doesn't work for the SMB. What works for the SMB doesn't work for service …
I can only offer a view from the standpoint of a "user" of OpenStack - I haven't read up on it nor implemented it. However our co-lo, datacentre and connectivity partner have been deploying OpenStack for a good 6 months + and we had an urgent need to deploy a Windows VM with 2Tb of storage as we've ran out of capacity on our co-lo SAN.
They had just announced their OpenStack platform as being production ready, so we looked at building a Windows 2008 R2 VM with Exchange 2010 on it as part of an existing DAG.
Way, way more trouble than it's worth. Now my basic / limited understanding of the platform does cause me to think that OpenStack runs "services" rather than VM's - similar to Microsoft's Private Cloud using SCVMM - so running an Exchange VM probably isn't a fair test.
However from an "end-users" point of view the entire experience was dire. Utterly terrible. Some bits that stand out:
- Terrible UI. Basic functionality in the web UI, rest is via shell
- Very complex. The open source boffins running it were in over their heads and struggle to understand it
- Doesn't play nice with Windows guests. At all. Drivers were a bloody nightmare
- Modularity is nice - powerful, but also brings with it complexity
- Performance was poor using Ceph. Although the 3rd party says this is due to a defragmentation that was running at the time. Regardless the disk IOPS are much worse on the VM on OpenStack with Ceph than a Hyper-V host and a Lefthand SAN.
- The overall experience I got had me going away thinking it's very immature
It's an interesting project, but the limited experience I've had with it so far leaves quite a bit to be desired. Lots of potential, and I'm sure if you've got dozens of open source experts to hand, lots of time and of course the budget for the hardware then you'd be laughing... although you could say that with nearly any "private cloud" or modern day "infrastructure platform".
Feels like it needs a couple of years to grow before I'd consider looking at it seriously for medium sized organisations.
Terrible UI. Basic functionality in the web UI, rest is via shell
Default UI is awful. Some of the commercial replacements are quite good. OpenNebula can be beaten into shape with enough effort.
Very complex. The open source boffins running it were in over their heads and struggle to understand it
No question. Anyone who tries to use Openstack without using a commercial distribution is in for a real headache. Anyone looking to use Openstack to "save money" is probably in for a real surprise. Pay the fee, get a commercial distribution and a lot of the problems go away.
Modularity is nice - powerful, but also brings with it complexity
Proof? In my experience swapping out components in Openstack has been pretty easy.
Performance was poor using Ceph. Although the 3rd party says this is due to a defragmentation that was running at the time. Regardless the disk IOPS are much worse on the VM on OpenStack with Ceph than a Hyper-V host and a Lefthand SAN.
Ceph is shite. Right terrible shite. But just about every storage widget on earth has a Cinder driver, and many have Swift drivers. Openstack screams when you use a Tintri as your Cinder storage! Maxta, SimpliVity or Nutanix make great hyperconverged solutions for Openstack, and I really like Nexenta Edge for Swift.
The overall experience I got had me going away thinking it's very immature
It sounds to me like someone deployed (badly) the reference implementations of the various modules, then tried to say "here you go, it's Openstack". That's like someone installing a basic Gentoo Linux and saying "here's your Linux server!"
Commercial versions make life a lot easier and have all the sharp edges smoothed off. Just like Linux can be greatly improved by simply adding in Webmin, Openstack works a hell of a lot better if you put a proper UI on the thing, replace Neutron's NFV with something that isn't awful and toss out Ceph for pretty much anything else.
Openstack isn't a product. It's a framework. The reference implementations are never going to be as good as the commercial products that slot into the various pieces of that framework.
And really, that's kind of the point. The interchangeability. If you want "free" then you'll get your money's worth. But if you are willing to pay for commercial products, but with the escape hatch that any one of them can be replaced if the vendor pisses you off, Openstack is the right choice.
Well, I'm not going to lie to you. If you try to use Openstack by going out and assembling the various open source components and trying to build up a distribution yourself you are going to have a terrible experience. The downside to openstack being so modular is that it doesn't come with a nice, simple installer, great UI and all the things that make it easy to consume gratis.
In fact, I'll be perfectly blunt and say that without using a commercial distribution, Openstack is more diffuclt to install and configure than System Center and about as friendly to manage and maintain.
That said, many of the commercial distributions have free or cheap versions/programs for home/lab use.
This is so bad from a major vendor's perspective ("What do you mean, they can just replace our stuff??) that I find it hard to believe it will get serious (deep) support. Sure, they'll offer a driver so they can check the "Supports OpenStack" box, but how can this be in their interest over the long term?
The vendor sales model is built around lock-in. The very last thing they want is to enable competition, so if you think the marketing guys are dictating engineering decisions today, just let them get one whiff of this and see what happens.
What are you on Trevor? Openstack is most certainly NOT ready. For it to work right you still need significant in house expertise (far more than most any other platform), large enterprises that can dedicate tens to hundreds of people can certainly probably be pretty successful(I heard ebay/paypal had something like 200+ people working on it for example), or outsource Openstack stuff to someone like HP and let them manage it(they have services that will do everything from provisioning to ongoing management - still even with them I wouldn't trust it with my mission critical stuff, the apps I manage aren't able to operate in something that is "built to fail" like a public cloud for example).
Everything I have seen and heard from people using it screams it is not ready, and won't be for some time(where time is easily 1-2 years, probably more). Just because you can plug it in and have it "work" doesn't mean it will WORK. Note that *my* standard for WORK is quite high.
Lots of organizations want people to think it's ready, and if you're ready to either really get your hands dirty (& take on pretty significant risk), and/or devote a good amount of resources to keeping it running then people best steer clear.
Last year I heard a great quote from one of the openstack experts at HP, he said something along the lines of "getting it up and going is easy, keeping it RUNNING is hard".
My former boss works with openstack on a daily basis at a large(r) enterprise (he knows my level for quality) and every time I see him he tells me it's not ready.
Openstack isn't ready if what you want it to be is an exact copy of the infrastructure you have today, but free. It will probably never be that.
But in terms of standing up a reliable, working infrastructure that behaves in the way Openstack was designed? It takes a couple of hours tops to get going and the commercially supported distributions are stable through every test I've been able to throw at them.
I have spent the past several months trying very - very - hard to break commercial Openstack distros through any sort of regular operation (including using automated configuration and deployment tools) and they are actually quite solid. It sustains failures, misconfigurations and all sorts of other bizarreness.
But yes, Nate, you are absolutely correct in that Openstack is not by itself a replacement for VMware in running your traditional legacy applications. That's one of the reasons you can manage ESXi with Openstack. You can also manage traditional metal servers in the same fashion.
Openstack isn't all things to all people for all workloads. But it is damned good at what it does, and you can bring your old workloads into the Openstack fold and have a single point of management.
Maybe you should spend some time with it and find out for yourself that it really isn't anywhere near as scary as you think...and that what you are expecting it to be and what it is designed to be aren't at all the same thing.
Take Openstack for what it is, not what you want it to be, and you'll find it's actually pretty good.
"it is a framework for doing IT infrastructure - all IT infrastructure - in as interchangeable and interoperable a way as we are ever likely to know how."
ER no......
It lacks some significant pieces, which may or may not be the fault of OS.
Fine get your cloudy stuff up and running, but on what & how?
That is the crux of the matter PHYSICAL hardware...., it's all very well being all cloudy and stary eyed, and building your software networks & systems, but there is an underlying PHYSICAL layer of networking & computing that is not handled by OS.
so no... it DOES NOT do ALL IT infrastructure and presenting it in a way that it does is totally misleading.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
