[HN Gopher] Red Hat Woos VMware Shops with OpenShift Virtualizat...
       ___________________________________________________________________
        
       Red Hat Woos VMware Shops with OpenShift Virtualization Engine
        
       Author : rbanffy
       Score  : 102 points
       Date   : 2025-01-16 14:45 UTC (8 hours ago)
        
 (HTM) web link (www.nextplatform.com)
 (TXT) w3m dump (www.nextplatform.com)
        
       | lenerdenator wrote:
       | What're the FLOSS technologies underlying OpenShift? Sounds like
       | KVM?
        
         | samcat116 wrote:
         | Probably kubevirt if I had to guess (which would then use KVM
         | under the hood)
        
           | houseofzeus wrote:
           | Yes you are correct, it is kubevirt and leveraging KVM as the
           | hypervisor coupled with the QEMU/Libvirt userspace pieces.
        
         | natebc wrote:
         | Openshift is Kubernetes+.
         | 
         | Openshift Virtualization Engine is kubevirt (aka kvm/libvirt).
        
         | lyarwood wrote:
         | https://github.com/kubevirt specifically for OpenShift
         | Virtualization
        
         | whalesalad wrote:
         | everything ultimately boils down to kvm or qemu, usually
        
       | antithesis-nl wrote:
       | Well, the Enterprise-hypervisor market is certainly _wide_ open
       | right now, due to Broadcom turning the financial screws on VMware
       | customers _hard_. There are, broadly, two categories of
       | potentially-profitable VMware customers up for grabs:
       | 
       | -Large enterprises that previously purchased hardware-with-
       | accompanying-VMware-licenses from OEMs like Dell-EMC: Broadcom
       | refused to even honor pre-acquisition license keys from these
       | sources, leaving many private data centers in the lurch, unless
       | they paid a _huge_ premium for a new Broadcom-originated annual
       | subscription (whereas the original key was one-off)
       | 
       | -Service providers with an ongoing "small-percentage-of revenue
       | per year, payable in arrears" agreement, that were suddenly
       | forced into a "hard vCPU and vRAM limit" subscription, payable
       | for at least 2 years upfront.
       | 
       | However, the magic word for both customer segments is "vMotion",
       | i.e. live-migration of VMs across disparate storage. No OSS
       | and/or commercial (including Hyper-V) solution is able to truly
       | match what VMware could (and can, at the right price) do in that
       | space...
        
         | zellyn wrote:
         | I think Oxide's stuff does live migration, but I don't know how
         | well it runs on non-Oxide hardware.
        
           | antithesis-nl wrote:
           | As far as I can tell, Oxide is all about compute, not storage
           | -- i.e., as long as you can guarantee a stable storage layer,
           | your Oxide racks will be fine.
           | 
           | VMware used to go a bit further, in that they allowed your
           | compute nodes to fail _and /or_ your storage nodes to fail,
           | without adverse effects.
           | 
           | If Oxide can do that, out-of-the-box, right-now, they'll be
           | having a field day. Otherwise, my reservations about Oxide's
           | business model remain...
        
             | zellyn wrote:
             | I _think_ their storage layer is fault tolerant. Not sure
             | though. Hopefully one of them will weigh in here.
        
               | steveklabnik wrote:
               | This is not my area of expertise, so I'll quote from our
               | website: https://oxide.computer/product/storage
               | 
               | The storage service uses OpenZFS for all data storage.
               | This marries Oxide's distributed data storage and multi-
               | node failure resiliency with the dependability and
               | efficiency OpenZFS has earned in its 20 years of running
               | demanding workloads.
               | 
               | The Oxide control plane monitors performance metrics as
               | another early signal of component failure. As sleds and
               | SSDs are rotated in and out, the Oxide control plane
               | migrates storage regions to ensure the appropriate
               | redundancy.
               | 
               | OpenZFS checksums and scrubs all data for early failure
               | detection. Virtual disks constantly validate the
               | integrity of your data, correcting failures as soon as
               | they are discovered.
        
               | zellyn wrote:
               | I figured ZFS would do the low-level redundancy, but I
               | was under the impression that Crucible does the higher-
               | level stuff, and I don't know much about it.
        
           | houseofzeus wrote:
           | I think most of these solutions including OpenShift
           | Virtualization, Hyper-V, Proxmox, etc. do live migration.
           | What the previous post is talking about is some of the more
           | advanced VMware live migration features like storage live
           | migration and cross-cluster live migration and some of the
           | automations layered over the top of them.
        
         | wesapien wrote:
         | vMotion is about host failure. If a blade server dies, the vms
         | can be spun up on another blade. Storage is same.
        
           | antithesis-nl wrote:
           | > vMotion is about host failure
           | 
           | No? It's also about load balancing and draining
           | compute/storage resources in preparation for maintenance.
           | 
           | Most pertinently: as long as your alternative doesn't cover
           | _any_ vMotion use-cases, customers will remain  'in talks'
           | with Broadcom...
        
             | toredash wrote:
             | Fun that people -still- don't understand the feature set
             | that VMware provides.
             | 
             | He/She is thinking about VMware Fault Tolerance
        
           | blown_gasket wrote:
           | You're thinking of high-availability, which registers the vmx
           | file of the VM to a physical server that isn't dead, then
           | powers the VM back on. Whereas vMotion is either a cold or
           | live-migration of the VM + memory state (if the VM is powered
           | down, there is no memory state to migrate).
        
         | trebligdivad wrote:
         | libvirt+qemu can do live migration across disparate storage; I
         | know quite a lot of people want to use it. I'm curious in which
         | ways you find the vMotion stuff does well with that.
        
         | noja wrote:
         | What can VMware do in this vmotion space?
         | 
         | The docs says open source can do a live migration, see
         | https://www.linux-kvm.org/page/Migration and
         | https://docs.redhat.com/en/documentation/red_hat_enterprise_...
        
           | mjevans wrote:
           | Vaguely like MS Active Directory vs Kerberos. The really big
           | thing 'vmotion' provides is 'things just work' vs flexibility
           | of options but more effort required.
        
         | SteveNuts wrote:
         | Hashicorp has some Nomad drivers for Qemu and a beta version
         | that uses libvirt. However it's fairly immature and lacks a lot
         | of features they would need to be competitive there.
         | 
         | Since IBM already has OpenShift I'm not sure how much time and
         | effort they want to put into Nomad virtualization, but I'd love
         | it as an alternative to Kubernetes.
        
         | tart-lemonade wrote:
         | Third is academia. Even with a marginal academic discount, it
         | doesn't come close to offsetting the price hikes. When the
         | choice is between slashing personnel budget and minimizing the
         | usage of/completely migrating away from VMware before the next
         | renewal, there's really no choice in the matter, especially
         | with the difficulty of firing employees.
        
         | lenerdenator wrote:
         | > However, the magic word for both customer segments is
         | "vMotion", i.e. live-migration of VMs across disparate storage.
         | No OSS and/or commercial (including Hyper-V) solution is able
         | to truly match what VMware could (and can, at the right price)
         | do in that space...
         | 
         | Someone's gonna start working on that soon. Necessity is the
         | mother of invention.
         | 
         | To me, this will be the UNIX wars moment for virtualization.
         | 
         | Originally, UNIX was something AT&T/Bell Labs mainly used for
         | their own purposes. Then people wanted to use it for
         | themselves. AT&T cooked up some insane price (like $20k _in
         | 1980s money_ ) for the license for System V. That competed with
         | the BSDs for a while. Then, some nerd in a college office in
         | Finland contributed his kernel to the GNU project. The rest is
         | history.
         | 
         | UNIX itself is somewhat of a niche today, with the vast
         | majority of former use cases absorbed by GNU/Linux.
         | 
         | This feels like an effort by Broadcom to suck up _all_ of the
         | money in the VMWare customer base, thinking it 's too much of a
         | pain in the ass to migrate off of their wares. In some
         | circumstances, they're not wrong, but there's going to be teams
         | at companies talking about how to show VMWare the door
         | permanently as a result of this.
         | 
         | Whether Broadcom is right that they can turn a profit on the
         | acquisition with the remaining install base remains to be seen.
        
           | pjmlp wrote:
           | I consider UNIX/POSIX becoming niche, including GNU/Linux,
           | for the so called cloud native workloads.
           | 
           | The large majority of managed languages being used in such
           | scenarios, compiled to native or VM based, have rich
           | ecosystems that abstract the underlying platform.
           | 
           | Moreso, if going deep into serverless, chiseled containers,
           | unikernel style, or similar technologies.
           | 
           | Naturally there is still plenty of room for traditional UNIX
           | style workloads.
        
         | gerdesj wrote:
         | >However, the magic word for both customer segments is
         | "vMotion", ... . No OSS and/or commercial (including Hyper-V)
         | solution is able to truly match what VMware could (and can, at
         | the right price) do in that space...
         | 
         | Storage vMotion requires a hefty license (as does DVS and the
         | other useful things, such as containers). Proxmox does it all
         | out of the box at a very reasonable price point.
         | 
         | Hell, VMware wont even let you use LLDP until your pissing
         | money out of all orifices. You get CDP only for "free".
         | 
         | After 20+ years of being a VMware fanboi I am migrating all my
         | customers to Proxmox. I've had enough.
        
           | zozbot234 wrote:
           | > Proxmox does it all out of the box at a very reasonable
           | price point.
           | 
           | Live migration of containers via the CRIU featureset
           | (checkpoint+restore in userspace, which is now part of the
           | mainline Linux kernel) is also an interesting theoretical
           | possibility - AIUI the Kubernetes folks are at least thinking
           | about supporting it. (Live migration fits remarkably well
           | with containers since it requires comprehensive namespacing
           | of all system resources - abstracting away from any
           | dependence on the local machine - which is also how
           | containerization works to begin with.)
        
         | mvdwoord wrote:
         | I would think another issue for a lot of existing large VMware
         | customers is NSX (incl. dynamic firewalling), orchestration
         | (vRA/vRO), vROPS, and the gazillion integrations that have been
         | built with tight coupling at both ends. Swapping that out is
         | not an easy task, and the "digital transformations" some of the
         | available products are banking on take a lot of time in real
         | life.
        
         | UltraSane wrote:
         | " Broadcom refused to even honor pre-acquisition license keys
         | from these sources, leaving many private data centers in the
         | lurch"
         | 
         | How is this not contract violation?
        
           | bigstrat2003 wrote:
           | That was my question as well. This seems blatantly illegal.
        
             | jandrese wrote:
             | The courts will make that decision in 5 years or so, but it
             | doesn't help people who need their stuff working today.
        
           | andrewf wrote:
           | AT&T sued them. https://www.ciodive.com/news/broadcom-att-
           | vmware-settlement-...
        
           | plagiarist wrote:
           | Sometimes it turns out that contract law only applies to the
           | poorer of the parties engaged in a contract.
        
         | oneplane wrote:
         | Xen (and XenServer and XCP and the Citrix banded stuff) have
         | been able to live migrate VMs for a really long time.
         | Definitely not something special to VMWare.
         | 
         | Realistically, all the legacy workloads (those that are
         | singletons and can't be load-balanced, need an active GUI
         | session etc) are going to to be problems forever, even if you
         | keep VMWare around.
        
       | whalesalad wrote:
       | If I were building a datacenter today I would go with proxmox.
       | It's "just debian" under the hood and can be customized and
       | controlled a multitude of ways (UI, CLI on the box, Terraform,
       | API, etc)
        
         | throw0101c wrote:
         | > _If I were building a datacenter today I would go with
         | proxmox._
         | 
         | I use Proxmox as well in a small-ish deployment, but have also
         | heard good things with Xcp-ng.
         | 
         | At a previous job used OpenStack.
        
         | NexRebular wrote:
         | Or if you'd like to help preventing linux monoculturalization
         | of datacenters, MNX Triton or vanilla SmartOS are very good
         | options too.
        
           | whalesalad wrote:
           | The cool thing about proxmox is that it is - again - "just
           | debian" so there is really no vendor lock-in. Yes they do
           | have commercial support/update subscriptions but the
           | community offering is open (https://github.com/proxmox). So I
           | do not worry too much about lock-in or monoculturalization.
           | At the end of the day it is a wrapper around fundamental
           | components of Linux. They do not have any proprietary secret
           | sauce that would F you down the road.
           | 
           | Correction I see now that the projects you reference are
           | Solaris based. I am down with that cause too - but if you are
           | a BSD/Solaris shop expect to do a lot of things on your own.
           | The linux virtualization space is substantially larger (not
           | necessarily suggesting it is better...)
        
             | jclulow wrote:
             | As an aside: Solaris is something you can buy from Oracle,
             | which they forked from OpenSolaris 15 years ago. SmartOS is
             | a distribution of illumos, which also forked from the same
             | code 15 years ago. They have since diverged, in some areas
             | dramatically, so we (the illumos community) don't bill
             | ourselves as being Solaris based.
        
           | yjftsjthsd-h wrote:
           | My impression, strengthened by
           | https://wiki.smartos.org/managing-instances-with-
           | vmamd/#usin... , is that SmartOS preferentially operates at
           | the per-host scale, which is probably a disadvantage in a
           | datacenter setting. (I don't know enough about MNX Triton to
           | comment)
        
         | zozbot234 wrote:
         | Doesn't Proxmox use a separate kernel package compared to
         | Debian? That's kinda annoying because it ends up making the
         | distro a 'Frankendebian' at best. Even using an up-to-date
         | kernel from the stable backports repositories is a lot better
         | than that.
        
           | brirec wrote:
           | They use a slightly modified Ubuntu kernel
           | (https://github.com/proxmox/pve-kernel), with things like ZFS
           | added. They also really are good about using proper Debian
           | tooling, and so their kernel doesn't cause any weird
           | dependency issues.
           | 
           | Right now they install proxmox-kerne-6.8.12-6 by default
           | (using pseudo-packages called proxmox-default-kernel and
           | proxmox-kernel-6.8 pointing at it), and offer proxmox-
           | kernel-6.11.0-2 as an opt-in package (by installing proxmox-
           | kernel-6.11)
           | 
           | I've been using the latest opt-in kernels on all of my
           | Proxmox nodes for a few years now, and I've never had any
           | issues at all with that myself.
        
             | zozbot234 wrote:
             | > things like ZFS added
             | 
             | That's a big gotcha - ZFS is non-free so of course it
             | cannot be part of Debian proper. Hopefully we'll get
             | feature parity via Btrfs or Bcachefs at some point in the
             | future.
        
               | Lariscus wrote:
               | Thats incorrect, it is free software but incompatible
               | with the GPL.
        
               | yjftsjthsd-h wrote:
               | > ZFS is non-free so of course it cannot be part of
               | Debian proper
               | 
               | ZFS is under the CDDL which is a perfectly good free and
               | open-source software license, just some people view it as
               | incompatible with GPL (IANAL, but this is apparently
               | somewhat controversial; see the wikipedia page) so Debian
               | doesn't distribute ZFS .ko files for Linux in binary
               | form. They do, however, have an official package for
               | it[1], just using DKMS to compile it locally.
               | 
               | [0] https://en.wikipedia.org/wiki/Common_Development_and_
               | Distrib...
               | 
               | [1] https://packages.debian.org/sid/zfs-dkms
        
           | Maskawanian wrote:
           | It certainly has an optimized kernel for its use case. I
           | believe it also includes ZFS by default. I wouldn't be
           | surprised if the Proxmox developers would prefer to upstream
           | these defaults, but they likely would introduce regressions
           | for the common use case that Debian optimizes for.
           | 
           | Ultimately, I use Proxmox as a hardware hypervisor only, so I
           | don't mind that it uses its own kernel. Everything I run is
           | in its own VM, with its own kernel that is setup the way I
           | want.
        
         | samcat116 wrote:
         | I'd be worried how proxmox would scale past a few racks. The
         | bones are all good, but I'm not sure how much scale testing
         | their API layer has had.
        
         | INTPenis wrote:
         | I'm using it now, even paying for it, and it works very well
         | but I only have one wish. That they could distribute an image
         | based distro, or a slimmed down appliance ISO. It just seems
         | unnecessary to have an actual OS on each host. Specially since
         | I've been impressed by Talos and Openshift for a while.
        
           | whalesalad wrote:
           | https://pve.proxmox.com/wiki/Linux_Container
        
       | stego-tech wrote:
       | They've been hammering us pretty hard, especially some folks in
       | the leadership chain who worked with them before. While I have no
       | direct beef with the product, the reality is that our Enterprise
       | workload (and in fact, _most_ Enterprise IT workloads in my
       | experience) are VM-first, not container-first.
       | 
       | My research conclusion at the time was that, while OpenShift is a
       | great product worthy of consideration, it really only shines in
       | organizations that are heavily invested in microservices or
       | Kubernetes. If you (or more specifically, your vendors) haven't
       | migrated into that state, it's not worth it compared to a RHEL
       | server license and their KVM+Cockpit solution for bog standard
       | VMs.
        
         | woleium wrote:
         | or proxmox.. i was going to say for small shops, but i have
         | heard of some larger deployments recently.
        
       | more_corn wrote:
       | Moving off VMware (yay!) Onto open shift (crap.)
        
       | breakingcups wrote:
       | Red Hat just hit us with 300% price increases for OpenShift
       | across the board right after we went live in production after a
       | little over a year of implementation. The entire org is very,
       | very unhappy about it.
        
         | woleium wrote:
         | An IBM by any other name would smell the same?
        
       ___________________________________________________________________
       (page generated 2025-01-16 23:01 UTC)