VMware is now part of Broadcom
broadcom.com
broadcom.com
VMware under EMC $625M acquisition lasted ‘04-‘15
Dell acquires EMC for $58B in ‘15 which includes previously acquired VMware.
Now Dell is trying to balance their books and sells entire stake of VMWare in ‘21.
Broadcom now picks up the pieces of VMware with acquisition completed this year (‘23).
I wonder which corporate overlord will take it over in the next 4-5 years.
Maybe Oracle or MS will be the next to bag hold.
There will be no next. Broadcom will get blood from the stone, rest assured. They will continue to raise maintenance and licensing fees until they very last customer turns off their last ESXi box. If you think IBM and mainframe is bad, you've never lived with a technology that Broadcom has acquired.
And don't get me started on trying to get their chipsets working on Linux. Up until recently, it was nearly as bad as nvidia's garbage. Even now, I'm not sure if everything is well supported; I've avoided BRCM wifi like the plague for years now.
Cheap is nice, but I miss the days when chips did things you wanted.
Before that Broadcom was (and still is) one of the larger communication hardware companies, making chips for cell, wifi, bluetooth, ethernet and ARM SoCs, along with telecom and data center boxes that used these chips. You almost certainly have their hardware in some device you own. But they have always had difficulty competing with Qualcomm, arguably in part due to anticompetitive behavior from the later.
After the acquisition, they have been acquiring enterprise software companies to diversify, including CA Technologies (think Atlassian/Oracle of the mainframe world), Symantec, and now VMWare.
Given that CA already had the reputation of "where software goes to die", this is ... bad news for VMWare. (I'd heard that description from many in the tech field, including from a CEO who's previous venture had been acquired by CA.)
I also hadn't realised that Broadcom itself had originated at HP.
The company that would later become Broadcom Inc. was established in 1961 as HP Associates, a semiconductor products division of Hewlett-Packard.[12]
The division separated from Hewlett-Packard as part of the Agilent Technologies spinoff in 1999.[8][13]
<https://en.wikipedia.org/wiki/Broadcom>
Citing: <https://books.google.com/books?id=WAW5DwAAQBAJ&q=1961+broadc...>
Part of me thinks that corporate acquisitions and mergers should not be permitted to use prior names, and especially not to adopt the name of an earlier, non-ancestor parent.
See AT&T and Lufthansa for a couple of examples.
Its history goes back to the 1960's. It is a major semiconductor company that I would say, was part of the ecosystem that made computers and networks possible.
You haven't heard of them because they make too much of everything that makes the fabric of everyday computing.
It's like how you would never really know of the company who made the pipes for your toilet/plumbing.
And more recently the working progress which is Intuit's acquisition of MailChimp.
The programming languages? What's the story there, were they acquired?
I used to work for HPE on their OpenStack team, and getting support, even as an employee, basically required me to have a rolodex full of engineers all over the world. A tremendous amount of troubleshooting was just looking up who had committed code, then tracking them down when something blew up.
I had one friend super into Tekkit but frankly "automate things so you can automate those things better" games never appealed to me a ton.
Take a look at this: https://www.youtube.com/watch?v=rR8W-f9YhYA
As I understand it, web Outlook pioneered numerous browser innovations that we take for granted now.
It's so bad, we have a dedicated email HTML person.
That's such a low bar. I'm still on "classic" at work, and I can't imagine what they would have to do to make it worse.
Yeah. They have done a better-than-most job of trying to default to open (for all of their other failings)
As for helping divide society/ politics, have you paid attention to HN front page the last few days? Wave and wave of OpenAI posts, with huge amount of ignorant, uninformed, speculative, conspiracy driven, repetitive, divisive comments that got endlessly upvoted and promoted. That reflects who the HN audience is and what we are interested in. You may say that HN also helps divide society/politics. Yet you, and I, are still here.
hmm, why would people still use VMWare? Honest question. Maybe there are some licensing issues I'm not aware of. Isn't vbox open source? If not, wouldn't even things such as https://copy.sh/v86/ in the browser would do most virtualization trick now days?
Esxi is an enterprise product which engineers are familiar with, the cost of the product vs the cost of migration and upskilling in alternative tech should not be underestimated, particularly when it’s the thing that runs all your other things.
Edit: in case you’re not aware, many organisations still leverage on-prem virtualisation technologies - that might not be obvious to people that haven’t seen it.
The money making part of vmware licensing is the baremetal hypervisor ESXi, the competitors are xen or hyperv and the likes.
The lock-in part of vmware is the vsphere management software, that allows you to move VM's en mass from one baremetal machine to another baremetal machine, or allows you to nest vm's, etc. manage your entire fleet of vm's which could be thousands or hundreds of thousands of virtual machines, from one management interface.
it's basically docker except it's actual entire vm's being moved around. VMWare ESXi being a baremetal hypervisor means you can run different OS's on top of these vm's and imagine being able to move these VM's all running different OS's around in your ecosystem.
That's what people pay the big bucks to vmware for.
It would be very difficult to build the vsphere/esxi ecosystem with pure opensource tools (it's possible with Xen, etc) but you'd be right back at paying some vendor a massive amount of money for building, integrating, and supporting this kind of system. (Redhat will happily sell you something that approximates vmware's tools, for megabucks).
As an aside, the consumer "vmware" software that you install on your workstation is such a small portion of their business, they basically spend no money on fixing/upkeeping. Apple silicon support was in beta for a loooong time, and they don't actually care about their workstation product. ESXi makes the money.
1. vmotion + storage vmotion - you can live migrate a vm from one hypervisor host machine to another. you can also live migrate the underlying storage (good if you want to consolidate storage servers, rebalance disk load, etc). with some caveats, you can do all of this without any downtime in the vm. it's not just a simple suspend on one host, resume on another host. a memory snapshot is migrated while the vm is still running on the first host, and when the amount of dirty pages starts to converge, they flip the vm over to the new host. similar idea for storage vmotion.
2. fault tolerance - for single cpu vms, you can use vmware's record-replay technology to execute a secondary vm in a "shadow" mode which replicates all of the nondeterministic events across the network. if one hypervisor host dies, the other can take over with no downtime. this is great when you need to add HA for a legacy application.
3. vsan - generally you run these systems with some sort of shared storage (nfs or iscsi attached SAN, or something like that). a SAN can be really expensive and a single point of failure. vmware can create a "virtual san" from a cluster of your esxi hypervisor hosts. as you can imagine, it has all sorts of HA features and can rebalance workloads to improve performance.
there are more, but that's just a few interesting features.
But it has been solid otherwise, even with in-place upgrades over many years.
I haven't run proxmox so can not directly compare :)
In every scenario that we spec'd out vSAN for production use it came in at least two times as expensive as your average dual controller, HA capable, storage array.
vSAN pricing is absolute nonsense.
In edge deployments where rack space is tight it's actually a great solution if you only have a few U to work with and have a HA requirement for a legacy app as well.
See VM.pool_migrate and VM.migrate_send https://xapi-project.github.io/xen-api/classes/vm.html. Those features got introduced in Xenserver 4.0 (2007) and 6.1 (2012).
Disclaimer: I work at XenServer.
Actually, no, they apparently won't, because I tried and was told they are discontinuing that product. They are going full steam on their OpenShift (k8s) product though, and will be happy to inform you that you can run a VM on that, but if you press them you'll find that you're actually running a VM in a container, and that you have to specify it as a k8s deployment, so apparently you can't just throw a VM up like you can in vCenter.
For the desktop, VirtualBox isn't much better. At one point after Oracle purchased it, they were tracking your IP and if they found you were using their extension pack that provides essential capabilities like USB 3 support (only the core of VirtualBox is GPL) and it was for commercial use, they were hitting you up for a license [0].
[0] https://www.theregister.com/2019/10/04/oracle_virtualbox_mer...
Think servers with four digits of days of uptime, in terms of things you probably shouldn't do, but absolutely can.
Source: I’ve responded incidents on those guests. It’s terrifying, but hypervisor uptime is limited by power not OS choice.
These days Windows is stable if you don't touch it.
VMware is stable even if one of the disks is out of space, the fibre is flapping, some idiot misconfigured the switches, and the whole cluster has time out of sync... by seven hours.
That's not hyperbole, that's an actual cluster that I got given to look after. It was running like that for months, perfectly fine, with VMware HA just "taking care of things".
If you sneeze in the direction of a Windows Failover Cluster it'll... fail.
Everyone has heard of it, it's theoretically comparable to the likes of VMware/Parallels desktop hypervisors, and it's cheaper. But just like spam (compared to ham), it's also obvious to anyone that's used both that it's by far the worse option, and there's very few reasons to use it besides cost, while there are numerous reasons to use one of the better alternatives.
In general, performance is significantly better with other hypervisors, but in particular shared filesystem performance (ie sharing a directory from the host OS into the guest) and features are lacking on virtualbox.
It's the default provider used by vagrant, and even the vagrant team recommend using a different provider for "any serious work".
Of course one could argue there's an even higher percentage of inedible parts in the canned monstrosity because it's all inedible, and I wouldn't argue that view point is wrong.
Hams like that often go on sale around US Thanksgiving and Christmas, with prices for them around US$4.50-6.50/kg last week and probably as 'loss leaders' around US$2-3/kg the week before Christmas. Ham is a regular part of 'traditional Thanksgiving' but not as much as turkey, whereas the 'Christmas ham' is definitely a thing.
One. NEW. CRITICAL. Alarm.
VMware's revenue for 2022 is $13B, and net income about $1.8B. Trim sales department, remove duplication(HR, IT, etc), cut down development, remove many make-work projects that the middle management engages in, increase licensing/support cost. They will focus more on net income, financing costs for $69B will be taken care of by layoffs and other stuff.
Sure, without a doubt, there are exellent devs to be found worldwide.
But let's face the reality ...
Western corporates don't go outsourcing in India for the talent.
They go outsourcing in India because they want to wield the axe on what they perceive as the expense of Western developers.
So, if they're not willing to pay Western developers, they're not exactly going to be looking at the premium end of the Indian talent pool either are they !
Especially as a typical remuneration structure for senior management is that you are rewarded with a bonus based on the amount of cost-cutting you've achieved.
I beg to differ.
You're either ignorant or have no idea how many SW and HW you interact with regularly from big name companies, has been through the hands of skilled Indian or other offshore devs which while being cheaper than American devs, are in now way worse programmers, sometimes the opposite.
You seem to equate pay only with talent and skill, but that's rarely the case. Lower paid international devs aren't necessarily worse than their American counterparts and well paid American devs aren't always better than international talent.
Your pay is more a function of opportunities and supply/demand in your area rather then purely on your own skills. Otherwise explain me how the same international devs who struggle to crack 50k in their own developing country can suddenly score 200k+ the moment they're in the American labor market. They don't magically become 4x better coders the moment they cross the border.
I am also not disputing that there is a proportion of companies out there who go to India for all the best reasons, a genuine quest for talent.
But sadly, I think we have to accept the reality that such companies are very much in the minority. The reality, sadly, is that the majority of companies who outsource to India only do so for one reason. Aggressive cost-cutting.
Of course companies will optimize for the best bang for the buck labor the same way you optimize for the best bang for the buck products when you shop for something, that's how capitalism works, it's a two way street.
Now if the move to India, or anywhere else, lowers the development cost without lowering the product quality, why is that a problem?
I would understand this being an issue when offshoring causes the quality of the product to go to shit, but it often isn't the case anymore, so what's then problem here?
Unless of course you mean the problem is "THEY TOOK OUR JOBS!" which would be understandable but like I said, capitalism is a two way street.
Supply and demand. American companies needed "developers, developers, developers", and when the supply is very finite and at the same time the investment markets were flush with "dumb money" that begs for at least a pittance of ROI, so in the end they ended up in bidding wars.
So now when a developer from <insert poor country here> enters the US labor market, their market value increases simply because they are now adding to the supply side of the US labor market - them being in the US is the factor.
In India, because of different levels of pay, one can get trained for first few years by working for cheap companies. Once they hone skills, they go for companies that pay them top: companies even below WITCH -> WITCH -> American companies in India -> American companies in America; that's how the pipeline works there. In the states, this pipeline is completely gutted: every company wants rockstars or hires fresh grads from select schools.
Also, you're making it sound like the devs can do whatever they want and happened to deliver cool new features. As said, back then everything hinted at them just switching Workstation to maintenance mode and letting it slowly fade into irrelevance, so it wouldn't have mattered how brilliant that new team was.
I bet the problem is that they are too “enterprise” and couldn’t price it low enough. If it were too expensive it wouldn’t be competitive with big cloud managed offerings.
Really it’s just Hyper-V with extra “cloud management” bits (“Azure Arc” and friends), but it’s _relatively_ friendly to manage alongside your Azure cloud resources once you set it up.
Bottom line is, I really don’t think they’re worried about on-prem hypervisors competing with Azure!
I actually thought that vSphere offers some stuff in the hyperconverged area that Hyper-V doesn't, this is where I could see it competing with Azure. However it seems Hyper-V with Storage Spaces Direct seems to be pretty much on par, at least on paper.
So there's even less reasons why MS would be interested in VMware. Only maybe to get rid of a competitor.
Obviously if you’re running dozens of hosts and thousands of VMs, needing live migration and things like that, then it’d probably be missing a lot of features you’d want, but for smaller stuff it’s pretty crazy how well it works.
I really thought there was a strong play there to do a major private cloud play as part of Dell's 'come back story' and then nothing.
But maybe that's exactly the problem. For a container based solution you still need a hypervisor but if you invite hypervisor people to the table, to they really want to champion linux containers or do they want to try to reach a local maximum by squeezing all of the fat out of VMs.
Even the Java to an extent 'got it' more than Dell+VMWare. I abandoned Java as a platform right around the time Docker went from whispers in dark corners to a quiet ping on people's radars. Within a couple years of that, the JVM team had increased their level of effort to shrink the system footprint from what I would label bet-hedging to aggressively. You need a small JVM if you're going to pack five services and/or ten JVMs onto the same host. Initially J2EE imagined itself to be multitenant, and it did a poor job of re-implementing half of Erlang, poorly. Containers were clearly on their radar.
Source: Someone higher up in my department.
They exist on legacy vendor lockin, and will milk customers until there is nothing left, which will take decades or more
In university, I had access to Oracle databases, Oracle manuals, Oracle Linux, etc. Not through some special university approved lab set up - I could just go and download them. Even their acquired software like PeopleSoft etc.
I had NONE of that for DB2 or AIX, for example.
And their respective market share, I believe based on no evidence but strong belief, belies that strategy.
(disclaimer, I guess - I work for IBM, but ironically as an Oracle consultant... the early access really did work :-)
People do know that oracle has ALOT more software then their DB right?
I mean their DB is the least vendor lockin thing they sell.....
That was how Adobe ran for a looooong time, up until and including CS6. They didn't give a f..k about piracy beyond something that could trivially be circumvented by a keygen and a few well-placed /etc/hosts entries, and that was what made them the utterly dominant power in anything creative - people were used to years of working with Photoshop (a friend of mine started with photography at age 13!), and so they demanded from employers that they use Photoshop. Incidentally, that also was what kept Apple afloat before the iPod/iPhone days - Adobe stuff just worked fine on Apple hardware but was a nightmare on Windows, so people also demanded Macs for their work.
The advent of CC came once Adobe had achieved that lock-in and started milking its customers for all they were worth, and additionally they opened up a load of legitimate customers as well who didn't feel like dropping a few grand on Adobe stuff but who cares about 50$ a month?
But on the other hand, look at how that turned out :(
From my perspective, Microsoft doesn't want to deal with HyperV anymore, they want your machines up on Azure. I'd actively advise against HyperV simply because I don't see that Microsoft cares about on-premises.
† RCT == Changed Block Tracking for HyperV, basically faster backups by allowing the backup application to know exactly which blocks of the virtual disks have changed since the last backup and the backup application can do fast incrementals via this means.
but i simply don’t know why there is a difference
edit: if you search “hyperv rct bug” you will find forum posts fasts showing the timeline of this issue lasting years and many clients complaining that microsoft would not even acknowledge the issue publicly (or even in ms support cases) until very very recently
I don't understand orgs that deal with companies with a loooooonnnnnngggggg history of predatory pricing and sales shenanigans and not have active mitigation plans.
Who has Oracle that isn't actively planning going to Postgres or MariaDB? Of course I'd say the same thing for IBM in the 1970s, Microsoft in the 1990s, and AWS today.
VMWare obviously has a lot of the same.
virtualbox is pretty capable but I only view it as equivalent to vmware workstation.
For my specific use case, it's display responsiveness.
My main work machine is a ThinkPad X1 Extreme Gen 3. Our development environment is Ubuntu, so when I got the machine our IT guy had helpfully installed Ubuntu on it.
There were two major problems though. I could only get my AirPods to pair as headphones, not as a full headset with microphone. And worse, I couldn't get my triple-monitor setup to work at all. (The ThinkPad has a 15" 4K display, and I use two 24" 4K displays with it: one in landscape mode immediately above the ThinkPad display, another in portrait mode to the left.) I could only get two displays out of the three to come up.
I did like the hardware quite a lot - I've been a huge ThinkPad fan for 25 years. So I immediately bought a similar machine for personal use. It came with Windows, and both of the above items worked "out of the box".
So I looked at the bottom of the work machine and saw that it came with a Windows license. I downloaded the Windows 10 ISO from Lenovo and installed Windows on it, figuring I would run Ubuntu in a VM.
I tried VirtualBox first, and it worked, but the display wasn't smooth. For example, I often use the Windows key + left/right arrow to move a window to one side of the display or the other. Ubuntu does a "sliding" animation when you do this, but it looked like it was only refreshing the display every tenth of a second or so.
So I tried VMware and it was perfect. The display is just about as responsive as running Ubuntu on the bare metal - every transition and animation is perfectly smooth.
VMWare Workstation (desktop edition) is faster and more polished than VirtualBox.
I'd like to try it all on top of ProxMox one day.
ESXi and the VMWare products built on top of it (vCenter/vSphere) are not even comparable to VB, other than that they both can run virtual machines. vSphere can move running VMs between storage or compute hosts without interruption, can failover between storage or compute, can failover between networking outages (thanks to virtual switches and the ecosystem of hardware support around it), and provides a platform for additional third party add ons for automated backups and recovery. Not to mention easy role based SSO access. My entire university's infrastructure was virtualized on VMWare aside from a few domain controllers and the Netapp storage clusters it all ran on, and the equally large Linux/KVM infrastructure and the HPC datacenter that ran a bunch of other stuff for...reasons (higher ed is fun). And as an added bonus, desktop type 2 hypervisors like Workstation or Fusion integrate perfectly into it. I used to manage a dozen Windows and Linux VMs straight from VMWare Fusion on my Mac and still do at home in my little VMUG cluster.
It's like comparing a Chevy Spark to an aircraft carrier, except you built an entire medium to large sized organization's infrastructure on top of it. You can't just switch overnight unless you want to stop making money for a while. For most orgs who can justify the already steep price, moving away from VMWare onto something else will mean multiple years long projects requiring thousands of person-hours to complete, redundant efforts (as the old stuff can't just go away until the new stuff is battle tested), on top of probable hardware purchases since VMWare and its demands have shaped on-site datacenter spend, layout, and networking for years.
The actual VMs are the easiest part to move since they are just some virtual disks and a config file. It's all the other supporting stuff and high availability that need to configured and battle tested that will take forever. It's not something you can plan to do ahead of needing to do it because doing so would mean doing the same job twice for years for a bet that you can't just weather some higher costs for a year or two before you can move stuff onto cheaper platforms (and train/hire for expertise).
More to the point, I think VirtualBox is the equivalent, broadly, of VMWare Workstation/VMWare Player; which is a specific individual product that runs on desktop as type 2 hypervisor and a teeny tiny bit of the broader VMWare ecosystem (probably neglible part of their revenue).
I don't know if VirtualBox has any product or share in server/datacentre space? Whereas, VMware is absolutely positively huge. The core ESXi product, sure; but again the ecosystem around it, from vSPhere/vCenter to vRealize and Orchestrator and nsx and vSan and everything else, the management and automation flows are pretty well integrated (externally; I'm sure it's a acquired/developed mess internally as every other IT product ever:).
It's a bit like... I don't know, "what does Window Explorer have that File Commander doesn't"? It's a valid question that has a rational answer which is useful for limited use cases, but it misses the very very big forest (Windows and Office and Azure etc) for very minor trees inside of it
Hope that helps a bit?
It's also licensable individually. If you want to use the VirtualBox Extension Pack in a business environment, you need to buy a per-user license. It's only $40, but Oracle has a minimum order quantity of 100, so you're spending at least $4000.[1] i.e. in a business, for about 20-30 users, VMware Workstation is cheaper.
[1] https://shop.oracle.com/apex/f?p=dstore:product:257141221156...
Is it at all comparable to ESX?
I kind of assumed the Proxmox was just the free and cheap and bare bones option.
I have worked on VMWare stack in previous jobs. but I run proxmox at home now.
Free ESXi without VMware tools is somewhat harder actually. Still far better reliability.
I'm sure that SmartOS is great, but being based on Illumos, I'd be hesitant to switch. Illumos doesn't have the most expansive community or hardware support as far as I know.
By doing on-prem private installation of it [0]. Support contract is not necessary. We run illumos on HPe hardware from Gen6 to Gen10 and there are no issues. Even have GPU compute going via passthrough to ubuntu bhyve HVM.
I remember taking a quick hobbyist look at JoyentOS or something like that about five years ago, but I thought the project died... I'm shocked to hear it's still alive.
VMware was a pretty good steward from my limited perspective. Does anyone have any experience with successful open source projects under Broadcom? They don't seem to have a good track record with driver support, at the least...
interesting, which ones?
I know a few organizations and vendors (like Veeam) were looking to RHV to be a good replacement, the announcement to discontinue the product seems to catch everyone by surprise.
I am still hoping Veeam will add support for a good 3rd option, proxmox, XCP, direct KVM, something...
were being the key phrase. Timing is everything, they Announced they were shutting down RHEV. AFTER broadcom announced they were buying vmware, seems like a terrible move in that light given that many vmware customers will be looking for a replacement in the next couple of renewal cycles.
Seems odd.
I have no use for kubernetes, I will never use kubernetes, I do not want kubernetes anywhere near me.
Most vmware customers I suspect have the same feelings
Set up a k8s deployment and you're fiddling with deprecated APIs every couple months - if you don't pay close attention the whole deployment spec falls apart within two years.
The VM stuff works for the majority of companies - you can even sprinkle containers in fairly easily.
It's still a massive if mature market that needs some attention and care, it's a shame it's going to get squeezed and abused by broadcom.
I never got this religious approach to tech when it comes to work. I have my preferences but ultimately I'll do whatever pays well.
What are the alternatives?
Is everyone on Hyper V already? Does Citrix still exist?
Are SuSE or Red Hat offering an 'open source' alternative? Surely people aren't using Proxmox in production?
In all seriousness, I think it is a hot topic in Enterprise Architecture (tm) meeting rooms ever since the merger was announced (I know it was in ours). But even if you find an alternative, you need to move, and a lot of companies have a lot of hard to modernize workloads which are skillfully managed by a lot of VMware trained personnel. No easy task.
Lot's do, e.g., the Austrian domain registry:
https://www.proxmox.com/en/about/stories/story/nic-at
And many others (albeit the big ones aren't listed there, a bit harder to get real testimonials from them, and we do not pester everybody):
The oVirt community's most recent release is from last December, so I'm not sure whether that project is going to thrive now that Red Hat has largely stepped away. (Last blog update is also December 2022.)
I don't expect Ovirt to survive, the vast majority of development was RH.
"The market" seems to "deciding" that in-house virtualization will be insanely expensive, and otherwise you need to rent OCP.
Now they bolted on virtual machines onto their OpenShift container platform and are pushing that.
..and your answer is kvm? KVM in itself is enterprise? ..or was it in context to my response referring to Oxide as an enterprise alternative? Well, in that case I can tell that there are only advantages to bhyve as the code base of kvm fork is old and all focus shifted to bhyve which build with todays hardware in mind.
https://klarasystems.com/articles/virtualization-showdown-fr...
edit: also bhyve runs on SmartOS too, and I think Bryan is going to be more comfortable with a Solaris OS under the hood
That is incorrect. It's faster than KVM's virtio, not faster than KVM's NVMe. They didn't bother to test KVM's NVMe virtualization[0]. To quote your source: "As it turns out, bhyve/NVMe isn’t just faster than bhyve/VirtIO—it’s faster than KVM/VirtIO as well" They're comparing apples to oranges. Why would they do that?
"Meet the Author: Jim Salter"
Oh, that's why.
>Bhyve may be juvenile compared to KVM but it seems to have a better design
Likely not, debatable.
>better license.
Debatable, but I get it.
>also bhyve runs on SmartOS too, and I think Bryan is going to be more comfortable with a Solaris OS under the hood
Most definitely why. KVM uses a lot of Linux kernel specific interfaces and functions. Bhyve is easier to port due to this.
[0] https://qemu-project.gitlab.io/qemu/system/devices/nvme.html
Would love to hear a counter argument or evidence otherwise but my impression is there are very few successful openstack deployments out there and mostly it just is useful to provide some negotiating leverage when talking to your VMware rep.
YMMV, but I think one of the main drivers of OpenStack is that it can be modified, because it's open source.
IE, if you're Verizon / AT&T / T-Mobile, it's valuable to be able to modify the product to suit your needs. I did some work on an OpenStack project a couple of years ago where the client wanted to leverage some hardware that was absolutely cutting edge. Literally so new, that engineers from the hardware vendor were involved in tweaking and optimizing things.
Try doing that with VMWare; it's impossible. With OpenStack, there's nothing stopping you from modding the product to suit your needs.
Obviously, this works particularly well if you're running hundreds or thousands of servers and you can justify the investment. If I were only running a few dozen VMs, OpenStack is probably overkill.
> Would love to hear a counter argument or evidence otherwise but my impression is there are very few successful openstack deployments out there
I've been doing OpenStack continuously for 10+ years, and it's never been more popular than today.
It's bizarre how people think it's dead.
The biggest difference between OpenStack in 2013 and OpenStack in 2023 is that in 2013 there were a dozen different vendors promoting it, and in 2023 it's basically Canonical and RedHat.
The CEO of Canonical wrote an editorial years ago, basically arguing that the implosion that happened in 2017-ish, when Cisco/VMWare/HPE/Pistoncloud/etc threw in the towel would be good for OpenStack. And I think that was true.
It's very similar to what happened to Linux, remember how there was a period where there were 20+ distros vying for supremacy?
Lots of reasons:
- the Python 2-3 migration clusterfuck (which is when I got to play around with it) burned a lot of people
- the learning curve just to get it set up and running is absolutely and horribly insane. It's understandable given its history of being open-source and because of that loved by universities who could shoehorn their existing crap infrastructure/hardware/designs into it. But that makes setting it up very very tedious and annoying, because there's countless options in the depths of its configuration files that you all need to go over so you don't miss anything.
- related: the effort required to set up clusters is soooo much higher for OpenStack vs VMware. VMware ESXi? A day or two, including racking and wiring, and I got something to show to my boss. OpenStack? Talk about three weeks until the myriad of its services are running without crashing under your feet twice an hour. Been there, done that, for both.
- the amount of people and skillsets you need to keep an OpenStack production cluster alive are, I'd say, double the headcount required for the usual VMware+Cisco+Netapp "standard" environment.
- there's barely any commercial support for OpenStack. And that's not "just" the usual vendor support side, but also the management side... finding freelancers or staff that has experience with VMware+Cisco+Netapp is easy, there's tons of people and MSPs with certifications on the market, but for OpenStack? Whoops.
Basically, even a small shop can't go wrong with a basic VMware setup, but OpenStack just doesn't make financial sense unless you're either an university (where you can hand over parts of the ops and support to students, and that has large enough demand for homegrown QEMU-KVM libvirt setups just being Not Enough anymore) or a huge institution (ISP, hosting provider, telco, large multinational megacorp) that wants to save the fuckton of money to VMware for licenses and has enough scale that the headcount for ops staff + Python developers is cheaper than VMware licenses.
What also killed a lot of demand for OpenStack (and a lot of other on-prem) was the general availability of reasonably-good-enough cloud providers. Why invest into an OpenStack environment and all the effort associated, when you can just rent servers on AWS?
Yes, https://www.xenserver.com/ and French OSS derivative https://xcp-ng.org
Citrix is now a private company run by former Broadcom head of software who negotiated VMware acquisition, https://www.crn.com/news/cloud/citrix-tibco-new-ceo-tom-krau...
Disclaimer: I work here
VMware's true magic is with vCenter. While Microsoft has an equivalent to that (SC VMM), nobody seems to use it because it is virtually unusable. I've never seen a successful production cluster of VMM running.
As to why there seems to be such an increase in security patches, its like the quote from Willie Sutton. That's where the money is.. The largest target gets the most attention.
Everybody has bugs. While you may hide under the radar with using some lesser known things, don't fool yourself thinking that there aren't holes the hackers can weave their way through.
I wish an MBA can explain to me the value of rebranding and losing brand loyalty/familiarity.
Broadcom would surely rise license costs while at the same time disinvesting.
Also, what will happen with VMWare’s Kubernetes investments? I am guessing all of their open source work will cease to exist?
Off the top of my head they have:
- Carbon black for what used to be called antivirus but has metastasized into XDR, an all encompassing endpoint security/detection/response tool
- A whole automation framework (vRA/vRO) that companies use for automating deploying VMs and other stuff. Probably straightforward to replace with other automation frameworks but migrating existing playbooks will take time and expertise ($$$)
- A whole virtual desktop management suite (Horizon) including SaaS IDP/Mobile Device Management (Workspace One/VMware identity manager). Can probably patch together a replacement with stuff like Jamf, your SaaS IDP of choice, and Microsoft's hosted VDI but it won't be quick and again device migrations might suck
- Software defined networking with NSX. This could be difficult to quickly replace if you have a whole system built around automating network segmentation/management with it. It seems like your choice is either go with a network vendor and lose tight VM integration or go with a lower tier virtualization platform and get a bunch of hacked together Apache/Linux native stuff and no support
- SDWAN stuff. Probably relatively easy to replace unless you have a huge number of branches or edge devices that need to be physically updated by rolling trucks
- VMware cloud where ESXi/vSphere run on public clouds. To replace it you need to stand up a whole datacenter/colo hypervisor environment/network/storage and staff up to manage them
I could go on but suffice it to say I don't envy the enterprise management types having to consider untangling their dependencies on VMware right now.
EUC, Aria and Carbon Black are still up in the air.
I've even used the Linux client to virtualize physical boxes.
This is textbook management 101, and not just for software companies.
I was surprised that VMware kept the "CloudHealth" branding as long as they did, - it turned out that the brand actually had some cache that VMware wanted to hang on to. But it was formally dropped last year, in favor of "VMware Aria Cloud Cost Control", or some equally overdescribed thing.
1. XCP-ng is a bit more polished and stable than Proxmox. Both are supposedly level 1 hypervisors, but while XCP-ng separates the hypervisor from the rest of the OS, Proxmox bumps it all together. Meaning that if something goes wrong with other part of the base OS, it will affect Proxmox.
2. It's simpler to do hardware pass-through on Proxmox (there's a GUI for that), but XCP-ng is more stable. There were occasions in which I couldn't pass through a device on Proxmox, but it worked perfectly fine on XCP-ng.
3. Xen Orchestra is the new manager for XCP-ng, the old Windows utility (which was great and had more functionality) is now EOL. But the creators from Xen Orchestra are too focused on getting VMWare clusters to migrate to their product and fail to add much needed functionality.
4. XCP-ng doesn't allow you to start VMs at boot in a certain order.
5. XCP-ng doesn't auto-mount shares when they come online.
6. XCP-ng doesn't have vGPU functionality for Nvidia or for newer Intel GPUs.
7. XCP-ng makes it much easier to create clusters than Proxmox.
8. Proxmox has a big community behind, meaning you can do some nifty hacks like: install MacOS, install Synology NAS software, do vGPU unlock (don't confuse this with proper vGPU support).
Bottom line: - I really want to like XCP-ng, but I'm migrating the work machines to Proxmox. Points 4, 5 and 6 are a big no-go, and it became a hassle to live with them. Unfortunately the developers, although polite, show no interest in improving those areas.
But yes, they have got it..
but I can only guess....
Then you will probably use some operations layer on top of that (e.g. QubesOS and XCP-NG are only Xen; Proxmox is only KVM/containers; libvirtd does both). Then based on that you pick your actual host OS (in case it's not already set).
I have GPU passthrough of the iGPU on a Ryzen 4000U working with kvm/qemu+libvirtd on an Alpine base. So the hypervisor runs headless (booting in blind mode) and the guest OS takes over the iGPU entirely. The caveats are explained in that arch Wiki article, IIRC. I think the least obvious part for me was sourcing and loading of the microcode/firmware.
Also had success with hybrid graphics (Ryzen APU + AMD RX GPU with one passed through and one on host) on an otherwise similar hypervisor setup on Arch BTW.
Even more so than usual, the more recent and obscure your hardware, the higher chance you'll get to compare different kernel versions, incant random undocumented parameters you found in some forum comment, or even reach for out-of-tree patches to get running smoothly.
Disclaimer: I work here
In WSL2 you can run GUI applications as well.
Case in point: QEMU now has support[0] for live migrating VM's RAM (like vMotion) and disk images (like Storage vMotion). But I'm not aware of any publicly-released cluster orchestration layer that does it for you - so you may need to build your own.
Happy to be proven wrong.
[0] https://developers.redhat.com/blog/2015/03/24/live-migrating...
Too bad Broadcom will also destroy the good parts.