[HN Gopher] Euro-cloud provider Anexia moves 12,000 VMs off VMwa...
___________________________________________________________________
Euro-cloud provider Anexia moves 12,000 VMs off VMware to homebrew
KVM platform
Author : xbmcuser
Score : 223 points
Date : 2025-01-13 12:19 UTC (10 hours ago)
(HTM) web link (www.theregister.com)
(TXT) w3m dump (www.theregister.com)
| wkat4242 wrote:
| Good to see companies calling Broadcom's bluff!
| depr wrote:
| Seems like a good move and everything but 12,000 doesn't sound
| like that many VMs? Is that a lot of VMs?
| zamadatix wrote:
| In terms of VMware customers it isn't a ton of VM but not
| peanuts either. E.g. the last healthcare place I was at (single
| customer rather than cloud provider) had ~30k VMware VMs and we
| were still small fish compared to some others. I've heard of
| places 10x the VM count as this making the move post
| acquisition - albeit less publicly.
|
| I think the purpose of the article is to highlight companies
| like this are starting/continuing to migrate post-acquisition
| rather than this particular customer was impressively large and
| did so. Particularly with the bits about the relative cost
| increase even though the customer was willing to walk away if
| needed.
| wink wrote:
| Single customer 30k sounds a lot easier to migrate (or at
| least to schedule) than 12k with potentially 3k different
| customers (probably more like several dozens or a couple
| hundred)
| zamadatix wrote:
| Could be, lots of pros and cons to each scenario - doubt
| either are easy by any measure.
|
| E.g. for about 20% we didn't even have a single piece of
| documentation other than the server name for who might
| actually care about the VM going down for the migration we
| wanted to schedule. Let alone how to test the migration,
| when it is best to do it, what software was actually
| running on it, if it's actually managed/monitored
| integrated with/by other systems whoch need to be looked at
| too, or if it could just be shit down instead (yay
| healthcare mergers and acqs). Our migration was also to
| VMware from (mostly) Hyper-V at the time, so not as much
| custom tooling needed.
|
| On the flip side a cloud provider is going to have all of
| the owner contact info but no direct control of the guest
| OS to effect the change so the battle is more with trying
| to get the customers to care enough to do the migration
| with you but not be so bothered by it all they up and leave
| your hosting.Not exactly a walk in the park either.
|
| In either case - almost never the tech that's the hard part
| for sure :).
| aaronax wrote:
| Difficult to say. That could conceivably fit in one rack of 60
| compute nodes (1/2U size) at 200VMs per node, leaving 15U for
| networking and SAN. Maybe $100/year/VM (rough cost of a lower-
| end cloud VM like EC2, droplet, etc.) in that case so $1.2
| million per year cost.
|
| Or it could take 10 racks and $50 million per year.
| Symbiote wrote:
| The article says, right at the start,
|
| > Anexia was founded in 2006, is based in Austria, and
| provides cloud services from over 100 locations around the
| world by placing equipment in third party datacenters.
|
| From the company's homepage:
|
| > The founder and CEO of Anexia [...] recently acquired a
| small hydropower plant in Kammern in the Liesingtal region of
| Styria for a "significant seven-figure sum" - i.e. several
| million euros. The power plant on the River Liesing generates
| 600 KW of electricity, enough to cover a third of the
| electricity consumption of Anexia's Vienna data center
|
| so this seems to be a significant operation.
| quesomaster9000 wrote:
| While average power consumption per rack has been
| increasing fairly steadily over the past 10 years, the
| metric I currently use is abound 10kW per rack under
| reasonable to heavy load - that's about the same as a
| consumer electric shower.
|
| So, this is implying their Vienna data center has 180
| racks? With 60 being about a third, if we say each rack has
| 40 servers... that's ~7k servers total... which is a
| sizeable chunk of floor space, like 3000m^2, or... 40
| tennis courts?
|
| But yea, that a non-insignificant operation just for the
| Vienna data center.
| zipy124 wrote:
| Entirely depends on context, like asking how long is a piece of
| string. One VM could be 200 cores, or it could be 1 core. It
| could also be a kubernetes/docker worker as-well so one VM may
| be thousands of containers. Finally they could instead just be
| important VM's. You could imagine a small or medium company
| having maybe 4 VM's each for prod, staging, testing etc...
| letting CDN's handle scaling (with everything else running on
| local dev machines) and so that 12,000 could be an entire 3,000
| companies whole stacks.
| mrweasel wrote:
| That depends on your context. 12,000 VMs is enough to a fairly
| large chunk of a small countries healthcare infrastructure, if
| not all of it.
|
| It a pretty decent amount of VMs, but not close to being
| unmanageable. I think it's more down to how the rest of your
| infrastructure looks, if we're talking about ease of migration.
| neilv wrote:
| Did Broadcom ever justify price-gouging captive customers after
| acquiring VMware?
|
| ("Sorry, folks, poor banana harvest this year, so we have to pass
| on the skyrocketing cost of bits licensing, which runs on
| bananas"?)
| rad_gruchalski wrote:
| They spent money on the acquisition and they would like to see
| some back. Isn't that obvious?
| toomuchtodo wrote:
| There is a lot of road between "reasonable return on capital
| invested" and "we are going to rake you over the coals so
| badly people are going to publicly flee to not be captive to
| us."
| rad_gruchalski wrote:
| Well, it's Broadcom and they just spent the money so they
| want some back. It's like IBM: acquire and rake.
|
| Hey IBM employees, it doesn't matter how upset you get. It
| is what it is.
| burnte wrote:
| No one said it wasn't obvious, we're saying spending almost
| $70bn on a company and announcing that the only way you can
| make it back is by squeezing 10x more out of the big fish,
| that might not have been smart.
| rad_gruchalski wrote:
| Apparently they are known for that so I'm not sure why
| acting surprised.
| fred_is_fred wrote:
| Why would they need to justify it? Lots of enterprises are
| captive to VmWare. Their entire IT staff has been using it
| their whole careers.
| bityard wrote:
| The first thing Broadcom did after acquiring the company was
| announce that they were going to offload their small customers
| and increase revenue by raising prices on their big customers.
| There was lots of warning.
| neilv wrote:
| Weren't they saying, _Once you are captive to a Broadcom
| product, Broadcom will have no qualms about totally screwing
| you with bait &switch huge price jumps_?
|
| Isn't that an interesting message to be sending, when much of
| their business involves getting some very price-conscious
| companies (think electronics bill-of-materials) to build upon
| their products?
|
| Or do customers of other Broadcom products already not trust
| them an inch? So Broadcom wasn't costing themselves valuable
| reputation here?
| UltraSane wrote:
| Why would a for-profit company have to justify increasing
| prices?
| throw310822 wrote:
| A-nexia, does it mean "lack of connectivity"? Great name for a
| cloud provider :)
| szszrk wrote:
| It's close to Anoxia and Alexia, which are also "nice" choices
| to be known for.
| pmontra wrote:
| Or "to annex" [1], as in all your VMs (are) belong to us [2]
|
| [1] https://www.merriam-webster.com/dictionary/annex
|
| [2]
| https://en.wikipedia.org/wiki/All_your_base_are_belong_to_us
| kgeist wrote:
| Anexia means "patience, humility" in Ancient Greek. Maybe
| that's what they were after?
| mrtksn wrote:
| If the claims about AI capabilities are true, it's matter of time
| before all providers to switch to homegrown solutions.
| cruffle_duffle wrote:
| The funny part is you can convince an LLM any bullshit you
| propose is right. It will agree with anything you say unless
| you encourage it not to and even then...
|
| " _VMWare costs too much. Our use case is incredibly niche and
| nothing out there seems like it would fulfill our needs. Can
| you help me write a new stack using bash scripts and AWK?_ "
|
| " _What an exciting idea! Bash Scripting is a clever way to
| orchestrate 12,000 virtual machines. Here, let me provide you
| with a bunch of code to copy and paste without doing any
| requirements gathering whatsoever._
|
| <pages of random bash scripts>
|
| _These bash scripts will let you run 12,000 virtual machines.
| Let me know if you need any explanation!_ "
|
| " _I forgot to me to mention we need to provision these using a
| rest api. Also these are globally distributed and require
| extensive monitoring_ "
|
| " _Not a problem! REST is an excellent way to communicate with
| the platform. I won't bother asking any follow up questions or
| gathering any more requirements. Here is all the code you
| need:_ "
|
| ... and so on...
| mrtksn wrote:
| If you know what are you doing in general, I'm convinced that
| it can actually help you out but I had 0 success with doing
| it brainfuck(chatgpt claims that anything beyond hello world
| is too complicated haha).
|
| Just like with the image and video generators, I'm fishing to
| find evidence of machines building something useful - had
| very little success so far.
|
| If the claims are true and these huge investments are not
| done in vain, not too far in the future we should be able to
| tell AI to build the tools we need, for example Microsoft
| should be expecting to discontinue MS Office either because
| someone can tell AI to make them an Excel instead of paying a
| subscription or tell AI to just do the job they will do in
| Excel.
|
| It's very strange that they talk about replacing the
| developers with AI but somehow still sell software. Something
| doesn't add up.
| kube-system wrote:
| > It's very strange that they talk about replacing the
| developers with AI but somehow still sell software.
| Something doesn't add up.
|
| There's a simple answer. This is the "replacement" fallacy
| that is all too common when hypothesizing the impact of
| tooling. Tooling increases the productivity of workers.
| This "replaces" workers in that you will need fewer of them
| to do the same job, but can't replace all of them because
| zero times anything is zero.
|
| If you have 10 people shoveling, you can replace 9 workers
| of them with a backhoe. But a backhoe can't replace all 10
| workers.
| mrtksn wrote:
| The talk is literally about replacing developers,
| engineers, and scientist with AI. It is not advertised as
| a developer tool, it is advertised as a someone who has
| PhD and ideally we will have AGI soon. companies are not
| talking about making even greater products they're
| talking about cutting off developers through the same
| thing.
|
| for example, salesforce proudly announced that it's not
| hiring the developers this year because they are going to
| use AI. they don't intend to make even greater software
| with the same developers using AI, they intend to replace
| them.
|
| i'm getting the vibes that nobody at the top believes
| that AI actually can replace anybody. I find this
| interesting.
| kube-system wrote:
| Advertisers are perennially full of shit. Generative AI
| is a technology, and just like _literally every useful
| technology created since the dawn of man_ , it improves
| productivity.
|
| A box of literal rocks does not do anything on its own,
| even if we have tricked them into thinking even better
| than we did last year.
|
| Literally for hundreds of years have people been fear
| mongering that some new technology will replace workers.
| It never happens. Technology displaces the available
| distribution of jobs, but it doesn't eliminate the need
| for people... and it never will, because human jobs are a
| human construct.
|
| > salesforce proudly announced that it's not hiring the
| developers this year because they are going to use AI.
|
| Yeah, more bullshit. Translated, they really mean: "Hey
| our AI is so great that we don't have to hire developers
| anymore -- it's definitely not because we have 70k+
| employees and our payroll is way too big -- it's because
| we're so good at AI and you should totally buy our
| software #EinsteinAIRocks #SalesforceOhana"
| quesomaster9000 wrote:
| I agree, it's absolutely and entirely bullshit and is
| becoming an easy sloganism that you can bash around.
| Where "replacing developers, engineers, and scientist
| with AI" really is about 'doing more with less'... or...
| just 'doing more, faster?'.
|
| But it is genuinely replacing workers, in real terms. And
| more importantly it's replacing services, which in turn
| means jobs and revenue. So they too turn to electricity &
| automation, so they can do more... so they can remain
| competitive.
|
| So uh yea, I 100% agree with you there, and wish that
| more people saw straight through it like you do.
| homebrewer wrote:
| They only mention KVM, however KVM is a kernel API and not a full
| hypervisor, and can't do half the things they claim it does (like
| support for certain disk images). What hypervisor are they
| actually using, does anybody know? Probably qemu? It doesn't feel
| like a major enough company to be able to write its own like
| Google did.
| klooney wrote:
| They probably mean libvirtd/qemu, when people say "we used kvm"
| that's generally what they mean.
| freedomben wrote:
| They are most assuredly using qemu. It's been my experience
| that when most people say "KVM" they usually are referring to
| the whole libvirt ecosystem. I assume that's the case here. I
| think of it very analogously to how people say "linux" to refer
| to a full OS/distro such as Fedora. Though "linux" is just a
| kernel, it's become common parlance.
| blueflow wrote:
| KVM is not only the kernel side - the userspace side of KVM was
| implemented in a fork of qemu. That fork later got merged into
| mainline qemu, so qemu is the "original" implementation of a
| KVM hypervisor.
| bertman wrote:
| Netcup seem to use qemu[0], which is what Anexia will also use
| according to OP's article
|
| [0]https://old.reddit.com/r/VPS/comments/1ghz4fd/i_know_contabo
| ...
| freedomben wrote:
| I definitely don't want to miniminize the significance of this.
| This is huge!
|
| But, they did have some major benefits that most companies
| looking to do the same won't:
|
| > _Anexia therefore resolved to migrate, a choice made easier by
| its ownership of another hosting business called Netcup that ran
| on a KVM-based platform._
|
| > _The hosting company is also a big user of NetApp storage, so
| customer data was already stored in a resource independent of its
| VMware rig - any new VMs would just need to be pointed at
| existing volumes._
|
| Again, still a great accomplishment and an exciting milestone for
| them, but for people still stuck on VMWare that are looking to
| migrate, it's good to know about the above things.
| tw04 wrote:
| Other than the additional capital outlay, this shouldn't prove
| a hurdle to anyone looking to do similar. Migrating from VSAN
| to external storage like NetApp is a non-disruptive process.
| And frankly VSAN has always been pretty horrible, so you'll
| likely end up with better performance and storage efficiency
| when all is said and done.
| amluto wrote:
| I found this rather odd:
|
| > "We used to pay for VMware software one month in arrears," he
| said. "With Broadcom we had to pay a year in advance with a two-
| year contract."
|
| If your goal is to extract every possible cent from your existing
| customers, why would you also switch them from net 30 to
| requiring partial prepayment? VMware wants money in general but
| should not have a cash flow problem, and forcing a monster early
| payment seems like it will force customers to notice an immediate
| problem and make a choice instead of allowing themselves to be
| slowly and steadily ripped off.
|
| If I were a pointy-haired CEO committed to the multiply-pricing-
| by-five strategy, I would do my best to sweeten the deal: offer
| generous payment terms, give nice-sounding discounts for up front
| commitments, give very large discounts for nodes that haven't yet
| been leased to a customer, etc.
| shepardrtc wrote:
| Either their sales team absolutely screwed this up, or they
| just don't want to bother with the VMware platform anymore.
| Maybe both?
| nyrikki wrote:
| Broadcom aspires to the Oracle model, they said as much when
| they bought VMware.
| diggan wrote:
| Perhaps similar strategy as spammers employ today? Where they
| try to filter away people quickly who wouldn't fall for it when
| they need it to.
|
| So do something slightly outrageous today, so you filter away
| the ones that won't stick around for the future more outrageous
| changes.
| nolok wrote:
| Because they have twelve thousands vms and are themselves a
| provider that can't afford to have downtime for its customers.
|
| So the thinking here was probably "there is no way they can
| refuse to sign right now and destroy their business in the
| process, so we might as well take the cake and also force them
| to stay after so they don't leave in 11 months and 29 days".
|
| Turns out that thinking is wrong for that specific customers,
| but for how many did it work ?
| tgv wrote:
| Indeed, a sales rep might have dreamed of an extra big bonus.
| rkagerer wrote:
| That's not sales it's extortion
| xbmcuser wrote:
| VMware was taken over by a company whose business model
| is extortion. Ie take over a company with customer that
| have few or little alternatives then keep jacking up
| prices as high as they can.
| malux85 wrote:
| No it's "vendor lock-in"
|
| Wait no you're right, they are practically synonyms.
| bhouston wrote:
| For the sales guys involved it looked like a massive
| payday for him, one he could brag about for years.
| luma wrote:
| The guy's name is "Hock Tan", it goes all the way to the
| top. Greedy billionaire trying to squeeze the entire on-
| prem datacenter industry. Every single one of my VMware
| customers is either in the process of migrating off or
| developing the plan to do so. At least one of them would
| be in Broadcom's list of 600 key accounts that Broadcom
| thought they could turn the screws on. They somehow seem
| to have forgotten that MS had just bought a chunk of that
| org and instead of paying VMware, they are now exiting a
| few dozen datacenters to move everything to the cloud.
| This org was _highly_ cloud-resistant (for a handful of
| good reasons), but VMware forced their hand at exactly
| the wrong time.
|
| I believe this course of action for VMware is going to be
| taught in business schools in the future.
| belter wrote:
| The story is even worst than this. You can find an
| interview from him on YouTube post acquisition of VMWare.
| A business reporter naively ask him what is his strategy
| for the acquisition. The answer just shows there was no
| strategy, just, and I am paraphrasing: "I spent all this
| money has to be for something"
| tgv wrote:
| The company I work for experienced this. The SaaS
| solution we depended on suddenly got very pricey. New
| pricing model and all. The sales reps were completely
| inflexible. It got so uncomfortable that I got to develop
| a replacement. When we were (gradually) moving over, they
| lost interest and let us off with a mild increase, and
| from this year on, we won't be needing them at all.
| rurban wrote:
| My cousin is a VMware sales girl. She didn't like the
| Broadcom move at all. Customers are exiting right and left.
| And no new contracts at all. Game over
| mbreese wrote:
| Here's an alternative theory - and I have no idea if it is
| right. But, this might have happened this way because Anexia
| _only_ have 12,000 VMs and Broadcom wanted to get rid of the
| account. I don't know if Anexia was considered a large or
| mid-level customer for VMware. As other have mentioned
| here... there are other customers who have many more VMs on
| site.
| bhouston wrote:
| I think the original theory is right. I've seen it play out
| close up. Basically a sales guy thinks they have a client
| who is caught and they can basically can extort them for a
| ransome and they try to do it. Sometimes clients actually
| are not as caught as the sales guys think and this happens.
| The sales guy looks now like an idiot and this is a guide
| that other caught customers can follow.
| mihaaly wrote:
| I still don't know.
|
| Starving the milk cows (push customers into losses) is never
| a smart strategy for those living on milk cows. Sounds more
| like inceadibly stupid. Or short sighted parasitic (squeeze
| all then run with the heist).
| ratg13 wrote:
| In my experience, VMWare attempts to force this model on
| everyone using tactics like not giving quotes until the very
| last minute, forcing buyers into a "take it or leave it"
| decision.. thinking (rightly so) that it will work in their
| favor most of the time.
|
| It takes a lot of balls for a company to "leave it" right as
| their contract is expiring, and speaks to talent and experience
| on the customer side to be able to stand up to bullying, and be
| able to pull off such a large migration.
| dylan604 wrote:
| But it's also a great negotiating tactic for the buyer not
| the seller. This seller has been chasing this buyer for
| however much time, and then at the last second walks away
| from the deal. I've had the price of a car drop drastically
| by doing this. I can't imagine a software sales person and
| its managers not budging and just letting the deal walk away
| either.
| hdhdbebd wrote:
| But it's highly unlikely the buyer will walk away if their
| core business already depends on the product you licensed
| to them
| ToucanLoucan wrote:
| The lock in is strong with their product and they know it.
| Migrating hypervisors is a long and arduous process for any
| medium-to-small business, and I speak from experience: it took
| our small team about 2 months to move off of VMware about a
| decade ago, also because the price of support was simply
| unhinged from our perspective.
|
| They would be fools to not expect high attrition of smaller
| clients, but big businesses and government customers aren't
| going to change, or at least not nearly to the tune that
| smaller ones would, and a smaller pool of larger customers
| paying a higher price probably works pretty well to keep
| revenues up while letting them slash support staff without too
| much of a reduction in quality for those that are left.
|
| It was clear to me from the beginning that this price hike
| wasn't about cash flow, not particularly. Broadcom doesn't want
| vmware wasting money supporting small fish.
| KingOfCoders wrote:
| I had my license cost from another large company YoY increase
| 10x (hefty amount). Reason was new sales manager who wanted to
| shake max money. They make revenue and then move on.
| segasaturn wrote:
| Anybody who formalizes their "homebrew KVM platform" into a
| marketable hypervisor product is going to make a lot of money I
| suspect. Every IT department I know is scrambling to replace
| their VMWare stack ASAP, including very large ones.
| formerly_proven wrote:
| > Every IT department I know is scrambling to replace their
| VMWare stack ASAP, including very large ones.
|
| At least some of the big ones seem to just pay up. Probably
| because they / their MSP / their relationship with their MSP is
| so dysfunctional that they know migrating is a pipe dream.
| stackskipton wrote:
| As former DevOps who dealt with VMware, it's not relationship
| with MSP. It's just so many things plugin to VMware that
| migrating off of VMware is just difficult. Monitoring,
| Backups, Deployment and so forth are deeply integrated into
| VMware so companies just look at work involved getting off
| and go, never mind.
| dehrmann wrote:
| Most things that would have used VMware 15 years ago are using
| Kubernetes now. The things that aren't are probably looking at
| Proxmox or just KVM.
| intelVISA wrote:
| Soon.
| icedchai wrote:
| Is there a reason Proxmox wouldn't work for many of them?
| kube-system wrote:
| Proxmox is limited in the tooling available to manage larger
| deployments. Although they have very recently made some good
| progress in improving support for these use cases.
| guerby wrote:
| https://pve.proxmox.com/wiki/Proxmox_Datacenter_Manager_Roa
| d...
| kube-system wrote:
| Yep, that's the progress I'm referring to. I'd say this
| is pretty basic functionality (managing more than one
| cluster) for enterprise type deployments. Proxmox might
| do a lot, but it doesn't have feature parity with
| VMware's stuff.
| downrightmike wrote:
| Or just use https://opennebula.io/ based on KVM, still works
| with vmware if needed. ~20 year old project. It has legs
| raincom wrote:
| 12000 VMs = 800 to 1200 baremetals. Since KVM can run vmdk
| images, it is more of how to manage all these barementals and VMs
| from a central place.
| olavgg wrote:
| That depends, on a modern Epyc, you can easily run a few
| thousand VM's on one system.
| rcleveng wrote:
| The phrase "homebrew KVM platform" made me chuckle. All of the
| hyperscalers have a homebrew, aka proprietary platform using a
| hypervisor. AWS has Nitro which seems to have been based on KVM
| but likely quite different by now.
|
| If you are selling VMs to customers, I can't understand a good
| reason to use VMWare. The only reason would be if you are selling
| VMWare as a service.
| TheNewsIsHere wrote:
| If you're providing application services to customers who need
| to use an application that the vendor only supports in VMware,
| then you don't have a choice.
|
| Well, I suppose you do. But an EMR/EHR for instance is going to
| _need_ vendor support, which means requiring VMware even if
| you're not selling VMware itself as a service.
| monocasa wrote:
| > AWS has Nitro which seems to have been based on KVM but
| likely quite different by now.
|
| It still is relatively stock KVM on the CPU side of things.
| They've been upstreaming changes they need like lower overhead
| for emulating Xen's hypercall interface.
|
| Most of their special sauce is in the devices though, as those
| natively provide VM boundaries leaving the hypervisor to not
| have to manage all that much at runtime.
| dilyevsky wrote:
| The challenging bits are all outside of the KVM like the VPC
| networking that has to be implemented using some SDN (e.g
| OpenVSwitch), block devices, etc
|
| VMware had a solution for all of these natively and with
| support. Not using their hypervisors you have to manage a huge
| pile of OSS+proprietary integrations and actually have staff
| who truly understand how everything works down to the lowest
| level. Doable but probably above the pay grade for most
| smetj wrote:
| Congratulations Anexia, well done! Pulling this off means you
| have experienced leaders and excellent engineers.
| pcblues wrote:
| I heard first-hand years ago of a bug in VMWare that was _CPU-
| Level_ for the company that rolled it out in an effort to replace
| their PCs that had software locally installed. The failure would
| have been catastrophic for a company that had just replaced all
| its PCs. However, they hadn't shipped the old PCs away yet for
| destruction, and managed to keep the lights on by putting them
| all back out again until the problem was fixed.
| hd4 wrote:
| Hard to imagine how other similar companies wouldn't just follow
| suit even if the cost of staying with VMware wasn't prohibitive
| (which is unlikely for most cases, probably).
|
| Stupid business decisions should be punished by the customer
| walking out. Homebrew KVM is not that hard.
___________________________________________________________________
(page generated 2025-01-13 23:01 UTC)