Fighter_M

>They're gonna get weight under control with these things eventually, right? Please define “we”, because say, Mazda has promised that the next MX-5 will actually be lighter than the current gen they manufacture.

They are no Nutanix competitor, Scale is just a rounding error compared to Nutanix’s revenue.

I’ll start with a little heads-up, because this one definitely lives in the "insider chatter” bucket. There’s no press release, no official breadcrumbs, nothing you can just Google and point at. You kind of "it pays to know the people" or just be close enough to the circle to validate any of it, but from how you’re talking about their conferences, every year, and so on, I’d guess you can sanity-check at least parts of what I'm telling here or about to tell. So, the Scale story... Take it with a truckload of salt if you want, treat it like valley gossip, or go ping your network and see what lines up. At some point, when VMware started stepping on rakes with all the subscription-only and "lets triple the quotes" stuff, Scale’s CEO had what probably felt like a genius play, which is buy back VMware licenses, sometimes even take hardware off prospects’ hands, then flip those customers onto Scale’s stack plus new Scale-branded servers. Super-seductive pitch, and it really worked! Sales popped, and everyone loved the story, except… VMware licenses aren’t really transferable, and that customer hardware, it’s basically dead weight. Lab it, demo it, donate it, or just write it off and move on. So sure, revenue charts looked like a clean hockey stick, but CAC which is "Customer Acquisition Cost" was quietly going through the roof. For quite a while, the board and VCs were allegedly getting the curated version of the sales pitch, which is lots of growth, lots of momentum, very pretty slides, and not so much focus on the part where the economics didn’t actually make sense, but cash always tells the truth, and at some point the investors, including Morgan Stanley, did the math and that didn’t land well. Scale's CEO gets pushed out, there’s talk of legal action, CFO goes missing, and suddenly you’re in shutdown and fire-sale place. Company gets sold at a loss, something like 20M under only Morgan Stanley invested capital from what I hear, and liquidation preferences mean one party, and it's MS people, is "fine" while everyone else basically gets zeroed. Investors, shareholders, everyone! Then you look at the team and it tells the rest of the story, which is engineering is thinning out hard, CTO moves to a competitor, something-about-the-nodes, VP Sales shows up at DataCore, CRO lands at Stor-stinking-Magic and it goes and it goes. Do a quick LinkedIn sweep and it feels like a quiet evacuation, so whether you were a fan or not, the tune right now is pretty clear. Yes, they’re still around, but it doesn’t exactly feel like a long runway situation.

>That's against the Veeam Comunity Edition EULA in most enterprise environments, so no, nobody is doing that, since it would come with massive compliance risks. As long as you’re not building your business offering on top of Veeam Community Edition, and you’re not stepping into their revenue stream by managing it for third parties, you’re in the clear. There’s no restriction on business type, so using the Community Edition in production is perfectly fine. Ten VMs restriction is a joke, though, but that’s another story…

> Docker/Kubernetes Using it for my homelab, nothing else. Only saw it inside software development devisions in companies, never in real productive use. Lots of software is delivered as containers these days. It is actually quite hard not to notice.

> What are the main issues with s2d you are seeing? There is zero built-in health monitoring or self-healing. Running S2D feels like babysitting a two-year-old toddler that you cannot leave unattended.

> For the love of all that's holy if you use Hyper-V don't use Storage Spaces Direct. This is 100% accurate. I can’t even count how many folks bought into the S2D “it just works” story and later got steamrolled by a possibly misconfigured cluster. Possibly, because every time an S2D cluster blows up in someone’s face, a bunch of Microsoft zealots jump out of the blue and claim it happened because the person who set it up did everything left-handed.

> As part of this project we would be provided with new servers. This would be a 2-node S2D cluster. I’m sorry to hear that. This design is basically asking for trouble. >No option for Proxmox, a SAN/DAS, 3-Node S2D or Starwind which I think would all be a better option. Right, anyone on this list would run circles around S2D, so the real sanity-check question is whether there’s any chance to reconsider.

> And if the OP wanted to replicate in near real-time, what would you recommend for that? The best option is to rely on replication that’s built into the application or platform itself, think SQL Server Availability Groups, vSAN, and similar. But before even going there, OP really needs to sit down and define realistic RTO and RPO targets. In most real-world cases, he’’ll quickly discover that async or pseudo-sync options, like replicated ZFS snapshots or Hyper-V Replica, are more than good enough, while being significantly easier to manage and safer to run overall. > My understanding is that Ceph needs more than two, but I haven't actually used it. Your understanding isn’t correct. You absolutely can run Ceph with two OSD nodes, you just need to place a third, MON-only instance somewhere to maintain quorum. https://docs.ceph.com/en/reef/install/manual-deployment This concept isn’t really different from most two-node HA storage designs out there, including V9-style active-active DRBD setup. It’s active-passive DRBD, which avoids a witness for simplicity sacrificing stability, and active-active redundant heartbeat networks guys who can surf purely two nodes, but with caveats.

> real-time or near real-time replication with drbd, ceph, or similar DRBD is never really a solution, it’s part of the problem by itself. Ceph is fine if it’s managed properly, of course, but it’s massive overkill here. There are much simpler, native ways to solve this, please see my original reply to the OP.

Yes, that’ll work, but it’s kinda overcomplicated, IMHO. LUKS needs a single block device, so you do need some merge layer, but… mdadm --level=linear …is usually not the nicest one! The simpler/cleaner way is to create N files, attach them as loop devices, put LVM on top, create one LV, put LUKS on the LV. Easier to grow later, fewer mdadm quirks. If you really want md, that’s fine too, but linear gives you zero redundancy. If you care about safety, use md RAID1/10 under LUKS instead.

I understand. That appears to involve a significant amount of work.

What workaround do you use in this case now? For us, backup is absolutely critical. We can and do change hypervisor vendors, but backup stays the same.

Lack of a software only version people could deploy on their own servers and no Veeam support were major limiting factors. I get why they never wanted a software version. Performance wise, their vSAN equivalent was not the brightest bulb in the room, so they did not want customers running any rat races they knew they could not win, but skipping Veeam and going with stinking Acronis?! That was a real shame! We’ll, it is what it is now, and another one bites the dust…

Yes, this is expected. See, Pacemaker has no awareness of active I/O, user sessions, or open files… If your constraints say ‘prefer Node 1’, and stickiness is set to ‘0’, it will immediately pull the resource back as soon as Node 1 is up again! DRBD also won’t delay a promotion just because someone is writing on the Secondary node, Pacemaker will simply demote/stop resources on Node 2, killing whatever is running. What should you do? Increase stickiness, which is ‘default-resource-stickiness=100’ or or even higher, so resources stay on Node 2 after failover. Use a ban constraint that is lifted only after you manually clear it or after DRBD is fully resynced. Don’t rely on a ‘prefer=50’ constraint for master selection and use DRBD Master/Slave rules or manual promotion logic. And absolutely enable STONITH, otherwise you will hit split-brain guaranteed! I mean with DRBD you’ll get either way, sooner or later, but the way you configured everything now it’s just a disaster waiting to happen. Good news is, there’s no filesystem corruption as long as DRBD is clean, just interrupted writes, which is an obvious data loss, but no corruption. But if a node flaps without STONITH, then yes, corruption becomes a real risk.

> I view it as an expansion of an engine start/stop system. Because it basically is!

Weird… We’ve got the complete opposite situation here: CVM delivers about 80–85% of the IOPS their Windows version can.

Yes, stick is a dying breed… Unfortunately!

>You would be shocked at how many people, when they say "I need a bigger car" actually mean "I want to sit up higher." Well pal, you’ve actually got a point there! Hopefully Honda’s marketing team is lurking around :) P.S. Man, I really miss my S2000! Take good care of yours, it totally deserves it!

Honestly, I just don’t see the WR-V making it big in NA, because most people here want something bigger, roomier, and with a little more power, while WR-V just feels too small for the typical buyer. To me, it’s kinda like a modern throwback to those old Civics from the ‘80s that college students used to drive around campus. Cute, practical, and cheap to run, but not really what most families or commuters here are looking for. I could totally picture a college girl or guy picking one up as their first car, but for the mainstream market? I doubt it :)

> cephfs is an absolutely perfect fit for you. Ever tried doing a 3-way replica in Ceph with one copy sitting miles away over a dark fiber link?

If you don't need to mount the namespace from multiple places at once, just run ZFS on the source and use ZFS snapshot send/recv on the target. Dead simple, tons of folks do it daily. Tinker a bit further and you can even offload cold data using nested ZFS with dedup turned on, so only dehydrated content hits the wire, but that’s more for the hardcore crew with serious guts with painfully slow WAN links.

>One thing I always look for is good documentation. Who should I avoid? Microsoft’s Storage Spaces Direct documentation, or rather the complete lack of it, sucks hard. Those of us stuck dealing with S2D end up relying on Dell’s or Lenovo’s docs instead. It’s a total nightmare and one of the main reasons we avoid S2D like the plague.

> Anyone recently done a VMware to Nutanix migration? I've got a small environment that I'll be doing soon. Small setup? Proxmox is a no-brainer. Got a SAN and running bigger ops? Then check out Hyper-V. Only after that, maybe give Nutanix a look, but don’t expect it to be cheaper than VMware.

> Given the scale Proxmox should do you well. If they run Linux, it’s Proxmox. If they’re a Windows shop, it’s Hyper-V. VMware’s off the table, no matter what.

> I like 2-node baby S2D clusters for little stuff (up to 30 VMs). Using Mellanox NICs and PowerShell hydration, we haven't really had any issues in 5 years or so. It was a painful ride for us, especially when patching one host, taking it down for maintenance, and boom, the second one crashes at the same time. S2D has come a long way since Windows Server 2016, no doubt about that, but we still prefer to avoid it if we can.

Trying to repurpose your existing vSAN hardware for Storage Spaces Direct is a recipe for disaster. Unless your gear was from the very beginning spec’ed by a certified Microsoft partner who set it up and will babysit your S2D cluster for you, don’t even think about it. VMware vSAN and Microsoft S2D are two completely different experiences.

Rebadged ScaleIO is pretty much like Ceph in that regard. Unless you throw a ton of hardware at it, you're not getting any decent performance. Even a four-node cluster makes zero sense with Scale/PowerFlex, both performance-wise and cost-wise.

Xen folks totally missed their boat like… Ten years ago?

>Hyper-V is in no way being 'killed off'. What do you think runs Azure? The Hyper-V that powers Azure and the Hyper-V you find in your Azure Stack or local deployment are two very different beasts.

>oVirt being dropped by Red Hat is a sore spot, because it's so so much better than Proxmox. Oracle guys seem to be more and more serious about their oVirt fork out.

>Why is XCP-NG obsolete? Vates They’re way too small to move the needle.

Did you try updating your O/N infrastructure? At least once?

Wow!

> We are a Veeam shop as well. Scale/Veem native is in testing now, was due to be release Q3/Q4. It’s not comparable features wise compared to VMware and Hyper-V versions.

Hmm… We found Scale way too expensive compared to Hyper-V, which we basically got for free with our Datacenter licenses. We already had the server hardware, but Scale insisted on a refresh using their in-house gear, which we didn’t want.

> until their cluster manager is proven, many enterprises do not want to go this route. Support is the bigger issue. Have you tried reaching out on a weekend or working with one of their partner reps to resolve a problem?

> For those who’ve already made the switch how’s your alternative working out? It’s Hyper-V, and we’re evaluating Proxmox, but not there yet.

You could be onto something there!

>The only problem is that the program seems to be abandoned right now, and no updates. Didn’t Oracle pick it up from the grass?

>Didn't HPE recently release an enterprise grade hypervisor recently? You mean Morpheus Data they bought assets from after they went belly up? No, it has nothing to do with enterprise. Hell, even Proxmox got more features!

Get Proxmox, re-use your existing SAN.

We built a few of those, Veeam backup repositories and IoT sensors for monitoring soil humidity and wine fermentation telemetry data.

>The problem, at least for us,with Scale and Nutanix is that you can’t grow storage independently of compute. You can have storage-only nodes with Nutanix, not sure about Scale, though…

>What would be nice is a filesystem similar to VMFS. I bet it’s not gonna happen. Clustered file systems are extremely complex, and even much bigger players, yes, Microsoft, I’m looking at you, have failed to deliver similar functionality for years, despite desperately needing it.

They have been revving their marketing way too high recently! So many promises made already. https://www.crn.com/news/data-center/2025/scale-computing-ceo-on-broadcom-vmware-fallout-nutanix-competition-veeam

> Did you ever deal with storage at enterprise scale? You made my day! Dude… In Spanish, Proxmox sounds like ‘sin señor enterprise’, and Blockbridge hits the same way, no matter how you spin it. Enterprises don’t buy storage from startups.

In Spanish, Proxmox hits like ‘sin señor enterprise’, and so is blockbridge you’re pushing here regardless of what. Enterprises don’t buy storage from startups.

> Check out Blockbridge Why? There’s no free version, and they’re closed source.

> Good partner to work with and competitive pricing. So, you’re actually pushing these guys, right?