Hacker News Reader

Comment by pfarnsworth | Hacker News Reader

They would own the datacenter, theoretically. People could come to Dell, and Dell would supply all of the hardware, from the storage to the virtualization to the servers. With VMWare owning Nicira, I guess that also including the virtual networking as well. It could be a compelling story, except IBM has already talked about how datacenter sales have dropped, so the US at least is probably overcapacity at this point. Bad timing, but not unexpected for a company like Dell, some this knuckle-headed is expected more from HP.

The funny thing is, virtualization is dying. All of the large private datacenters these days are non-virtualized bare metal on commodity hardware. Virtualization is an unnecessary overhead when it comes to datacenters these days. And enterprise storage is also not "web scale" since it's faster to shard data across 100k servers than have a single huge database with a single point of failure.

View on HN

> All of the large private datacenters these days are non-virtualized bare metal on commodity hardware.

This is so hilariously wrong I can only assume it's from a marketing brochure.

Just one data point, but at Spotify we have almost all our stuff deployed in our data centers. ~10K machines, all bare metal, barely any virtualization, in a handful of sites.

It's just cheaper and easier for us to have a lot of hardware racked and stacked and then spin it up as needed without having to worry about virtualization. Experience for our devs is pretty much the same whether they're provisioning a bare metal machine as it would be for a VM.

Out of the SF bubble, the exact opposite is true. In Enterprise environments, racked servers running VMWare hosts still reigns supreme.

It basically is. It's almost exactly verbatim a statement made by Google when talking up their container game.

Doesn't Google make up a large portion of the private datacenter space?

> The funny thing is, virtualization is dying. All of the large private datacenters these days are non-virtualized bare metal on commodity hardware

This is wrong on so many levels. Virtualization is, and will continue to be, a integral part of datacenters. First and foremost, it enables you to deploy one single image to any machine you have, if you need computation nodes. Or it enables you to have multiple VMs on the same hardware. Both of those apply to commodity and server hardware a like.

I think "dying" is perhaps a bit of overstatement, but there is some merit to the notion.

Containers may change the game. If you can containerize on top of something like Mesos, Kubernetes, etc - there is no need to run on top of a virtualization layer.

Virtualization was a buzzword, and per se, used for plenty of things it shouldn't have been used for. But it will continue to be a cornerstone in any IT infastructure for the foreseeable future, at the very least until the containers mature to a point where it can replace virtualization because it has near bulletproof sandboxing.

> The funny thing is, virtualization is dying. All of the large private datacenters these days are non-virtualized bare metal on commodity hardware.

Well, you sound like a Google employee, because that's pretty much their party line.

Of course, reality doesn't mesh up with that when you step out and look at other data centers.

This is a very, very wrong, and unnecessarily harsh and personal in its wrongness, comment. The person you are replying to is dead on and I suspect you are conflating "I support a lot of virt environments in my profession" with the state of the industry.

At scale, in prod, virt is legacy. Aside from Linode where I obviously ran virt, the only virt I've ever touched in private datacenters is relegated to labs or testing farms. We have a lot of tech, both from supercomputing and the new valley stuff which is inexplicably rewriting all of that, that makes virtualization completely unnecessary outside of a multitenant situation with separate paying customers and security domains. Even there multiple vendors are working on it, notably Intel, who is pushing VT-x into containers with multiple efforts.

Don't be so confident to talk about reality, because yours is very different from mine. The original poster was unguarded with their claim but in terms of winds of the industry, they couldn't be more correct, and if you think I'm wrong you're on the wrong side of the shift that is coming, nigh already here.

Example: We bought space in Virginia for Foursquare and didn't deploy a single byte of virt. My current employer has dozens of facilities and virt is a lab thing. No prod, anywhere, across dozens of products and lines and organizations, uses virt. Google doesn't, right you are. Nor does Twitter, who is deep into Mesos (same for everyone who is also deep into Mesos or its many friends).

You might scoff and say well, my CIO says, and you'd be correct today, but the state of the art for resource utilization at scale moved away from virt because we figured out that running full operating systems next to each other as a bandaid for bad CD and provisioning and resource allocation stories is a shitload of overhead for zero gain. There is absolutely nothing that virt gives you, aside from a perf hit and unpredictable low-level behavior (some of us care about cache lines and context switches), which cannot be implemented with tooling atop bare metal platforms. Virt makes your hot aisle hotter so you don't have to figure out bare metal provisioning. You should care about that, then figure it out. It's not hard.

If you are strongly convinced that I'm wrong, much like your neighbor posters, the state of the art simply hasn't made it to you yet. Sorry. You should be willing to consider, however, instead of sniping at change like your tone will keep it at bay. There is a lot of denial in this thread, and it's trivial to deduce why that is.

> the state of the art for resource utilization at scale moved away from virt because we figured out that running full operating systems next to each other as a bandaid for bad CD and provisioning and resource allocation stories is a shitload of overhead for zero gain

And people are moving to OS level virtualization instead. But the point still stands, for any independent datacenter, there is plenty of business sense in using virtulization to serve customers needs. Heck, even for an internal datacenter, hw virtualization makes sense for developement, testing, and general infrastructure.

For application specific purposes virtulization was never really a good idea to begin with, plenty of people have said that ever since it became a thing, but "State of the art" was virtualization. Now the developers realise that virtualization wasn't the way to go, and revert back to bare metal, or OS virtualization. It is a classic example of a fad because it was a buzzword.

Amazon's virtualized scale is the counterpoint to bare metal being the only "state of the art".

Virtualization is not going anywhere, and I don't think most companies will be adopting bare metal en masse as you suggest especially as virtualized extensions to containers like rancherVM become commonplace.

Also, many in the financial sector prefer the security benefits of hypervisor and VT-x isolation to reduce exposure to kernel/hw-level exploits.

> This is a very, very wrong, and unnecessarily harsh and personal in its wrongness, comment.

In other words, you are ~very~ (edit:) Exceptionally offended that your opinion is different than mine from your experience, whereas most of the paperwork on the industry as a whole does not agree with your experience.

I get that you're mad, but you should take your emotions out of the equation and look at the numbers published by literally every industry analyst.

No. Your quote does not support your rewording in the slightest. You are interpreting me as angry and offended because it makes dismissing me easier, and because it is mirroring your own feelings, for what it's worth. It's completely irrelevant, but I am not malcontent at all even despite your downvoting me for a well-argued counterpoint.

Perhaps it is not me who should step back and reevaluate.

Part of the problem here is a lack of specificity on the industry. Since you invoked analysts, Gartner and the typical HN view of "industry" are wildly different, but I would posit what happens in what we typically call the "industry" is in the pipeline for the Gartner side in about a decade. However, tickers I would normally put in the Gartner/CIO bucket are aligning with me on this, more than you'd expect. Even Manhattan finance.

And yes, I am aware of CAGR forecasts for virt, but a big driver of the market's growth is expansion of virt deployment footholds thanks in no small part to momentum fueled by opinions like yours. There is also an incentive to sell virt by hardware and procurement vendors, because you need more fleet to do the same work under virt, unconditionally. The market will level off because fewer new projects and companies are reaching for virt as evidenced by, yes, Google, and half the other household names in the valley.

Thought exercise: Google published Dataflow out of their work on streaming architecture and said they are moving on from MapReduce (for the most part). If you got research that says the Hadoop market is growing, wouldn't you look at it objectively in context since the very organization who defined the technology has moved on from it? Market research and analysis often lacks frontline context, much as it does here.

How about some references, all I see are some assertions being thrown out there?

>Well, you sound like a Google employee, because that's pretty much their party line.

Do you have a citation for this or are you just propagating your personal issues with Google again?

>> The funny thing is, virtualization is dying.

I disagree, the amount of dependency we have on virtualization at my current organization ( which has been a leader in On premises software systems ) is huge. And we are still generating the majority of revenue from there.

I disagree with your second paragraph.

What I see Dell doing is consolidating products in a space that is shrinking while offering a unified datacenter product which has two target customers:

* Bringing dinosaurs into the modern age with a "private cloud" * Bringing maturing organizations into the physical realm offering the advantages and alternatives to public cloud offerings

The funny thing about HN is how easily people discount Azure and AWS as the two largest computing clouds on earth, all virtualized.

I think there is a great deal of ruin remaining to milk in the enterprise market, and VMware has little competition in most customers.

> It could be a compelling story, except IBM has already talked about how datacenter sales have dropped, so the US at least is probably overcapacity at this point.

Counterpoint: Dell's modular datacenter solution is doing well and really murdering the competition.

Reply on news.ycombinator.com