Opensource replacement for vSphere/ESXi?

AnthonyC · January 13, 2016, 11:37:09 PM

I am looking into the possibility of replacing vSphere/ESXi at work; I know there is Openstack with KVM but I think that's a bit too much. Was looking to try something like oVirt/RHEV. Does anyone has experiences with that?

wintermute000 · January 14, 2016, 12:04:47 AM

You don't have to run openstack? It's not a fair comparison with vsphere /esxi, it's an orchestrator I thought?
I would say proxmox + kvm = vcenter + vsphere?

AnthonyC · January 14, 2016, 08:17:56 AM

Quote from: wintermute000 on January 14, 2016, 12:04:47 AM
You don't have to run openstack? It's not a fair comparison with vsphere /esxi, it's an orchestrator I thought?
I would say proxmox + kvm = vcenter + vsphere?

Cool I would take a look at proxmox; I do like that ovirt is backed by RH. Also found a reddit on this
https://www.reddit.com/r/homelab/comments/2tv52i/help_ovirt_vs_proxmox/

wintermute000 · January 15, 2016, 07:11:47 PM

Thanks, looks more enterprisey and I cut my teeth on fedora core so I'm more comfortable with RH vs Debian base, looks worth checking out

burnyd · January 17, 2016, 12:15:57 PM

Openstack can pretty much orchestrate whatever. Its as powerful as you would like it. But I have read a lot of articles where things have went bad.. like really bad haha. I mean its like anything else where a group of users have full control to anything open source.

It really depends on your use case. Hyperv is making excellent progress and I feel its pretty much enterprise ready at this point looking at all the customers running it. If you have ELA with microsoft and you run at least DC version of windows then you pretty much get free virtualization.

If you have a group of developers and people who are completely comfortable for linux concepts then give KVM a shot. The problem there is there is no overall orchestration so you have to build something like openstack on top of it to orechestrate network,compute,storage and authentication.

AnthonyC · January 19, 2016, 03:29:22 PM

I've recently seen an Openstack implementation that was pretty impressive but it exceeds my requirements and will just add in unneeded complexity. Beside I am more interested in checking out Cloudstack since it is supposed to be easier to deploy.

At my current shop we do have developers who mainly code in BSD/Linux so I am interested in testing KVM with some sort of management software on top of it. This is a project that is probably at least a few weeks away though.

wintermute000 · January 19, 2016, 03:47:16 PM

Coincidentally this topic is coming up for a SMB class client of mine (three prod hosts + DR, around 8 multi-tenancies with 3-5 VMs in each). I'm steering him towards oVirt + KVM though Hyper-V would be the easy choice, but he's mindful that long term MS can turn around and start turning the scews on licensing just the same as vmware.

We've guestimmated that it would take 3 months(x2 for his senior techs) to execute a migration
- 1 month learning/fiddling
- 1 month PoC/migration planning and testing (live failover, rebuilds etc. so his techs are comfortable)
- 1 month migration

At that rate he's looking at several years before he makes ROI but as I pointed out to him, that assumes his vmware costs stay static - and if his business expands which is obviously his aim, he will only need more hosts, more features, more licensing etc. Whilst he's still relatively small and not bound to the enterprise plus level features and the hypervisors are just acting as commodity, the time to jump is now.

a major obstacle is that there is no easy migration path from Vmware to start using Ceph (he's still on DAS/local, ugh) without building a pair of redundant iSCSI/NFS target servers, which in his budget would pretty much push back the ROI (from dropping vmware licensing) a year or more. I'm telling him not to persist with local/DAS though, but as you can imagine at that size, budget is really tight. Though that is obviously an option. In an ideal world we would have a distributed, open source (free) shared storage platform that we could build out that supported both vmware and kvm environments natively (and yes pigs might fly LOL).

I'd be curious though if he does fold and go down the path of 'least resistance' (Hyper-V) esp. as majority of the client workloads are MS as well.

burnyd · January 22, 2016, 10:11:42 AM

I mean if he is that cheep that the VM's are using DAS then why even get something like vcenter/hyper-v because there will never be a live migration. I guess it all depends on how you build the application. If you front end everything by a virtual ip / some ghetto global sight load balancing you can do this for sure. I mean that is the way the world works anymore that the redundancy should be in the app stack not in the os if that makes any sense?

wintermute000 · January 22, 2016, 01:48:00 PM

Absolutely, but try telling your clients. I'm actually kicking off another layer two dci project shortly, sigh.

Even that Gartner report didn't work LOL

If they aren't willing to bite the bullet and start mandating scale out / cattle over pets etc not much you can do.
And for my smb client, his vms are his client's A.D. Exchange etc so he can't go cattle

burnyd · January 22, 2016, 10:35:02 PM

wtf? Why? vplex is expensive plus keep in mind they are going to have at least a vmax 40k at each site and they are crazy expensive. Do they have a VMware team? Considering that the vplex to do synchronous is prolly pretty low latency why didnt they just look at a stretched vsan cluster?

wintermute000 · January 23, 2016, 12:40:29 AM

No idea. by the time we got there the spec was in stone.
They are fairly large though yes in house vmware team.

At the end of the day server guys are just in love with vmotion because it makes their lives easier and as external consultants we have no say in the overall stack and just have to deal with it