Scale Computing is a virtual infrastructure company that sponsored Virtualization Field Day 4. The company, along with others, paid money to Gestalt IT to participate in the event. That money went towards funding my trip to the great city of Austin Texas from January 13-16, 2015. That money also paid for my food, lodging, and perhaps a few spirits. I received no compensation from Scale Computing, Gestalt, or anyone else to write this post, and none of the parties have viewed this post before it went live. Today, we continue our discussion with a grumpy, cynical, and almost angry Riker, my alter-ego, about Scale Computing’s product.
SRiker : Who from #VFD4 is next at bat?
Oh just a hyperconverged player in the Linux virtualization/SMB space that presented an hour or two before I boarded a jet and flew home. I hadn’t even heard of them before #VFD4…Scale Computing.
Scale’s based out of India…..napolis, Indiana, as the company’s Twitter account joked recently. This is hyperconvergence, Hoosier-Style.
Riker: And they do what exactly?
They’re in the hot hyperconverged market, but, interestingly, they run a forked KVM as hypervisor unlike the players I was more familiar with. So I got to revisit that Swedish guy Libvirt & his sidekick Qemu during this presentation.
Riker: What else stood out to you about the presentation?
One other thing about Scale stuck with me..what was it now..
Riker: Hoosier VMs?
Nah. I already made that joke.
Riker: And you got in the crack about funny Linux names…
I know. Damnit! It’s on the tip of my tongue…
Oh wait. I remember.
As Tech Field Day events go, this one really kicked ass! Not only did we geek out on virtualization stuff, but Scale brought some servers & switches in a crate that you could wheel around or maybe even ride, like a pony.
Also, the Scale CTO was fun; he wore cowboy boots & a beige blazer and at the start of the show, he was like, “Powerpoint?!?! What’s that?!?!” and then he was like “Demo Time!” and finally he was like “I wonder what would happen if we pulled power from a cluster node?” and then he did it Riker, he did it, cause he’s a baller and he has no fear!
Riker: Wow. You’re easy to impress. Calm down and tell me who they’re trying to sell to.
Well, unlike other appliance slingers, Scale Computing is focused entirely on IT shops in midmarket/small business.
Riker: They’re pitching hyperconvergence downmarket, where the only thing thinner than the margins are the skillsets? Is this even a viable market?
Hey! Knock it off with the “downmarket” conceit. Small businesses have IT needs too, and Scale thinks there’s still a lot of runway left for on-prem virtualization in businesses with 25-500 employees.
Some of the Delegates challenged Scale on this point, but our cowboy CTO knew his Dunn & Bradstreet data, saying that the Fortune 500 are well-served by incumbents, but no one is delivering hypercoverged systems to the 300,000+ “midmarket” businesses in America.
Riker: So? You and I have worked in such places. They’re ultra-cheap, or as you would call it, “value-oriented.” They don’t like complexity in IT systems because they usually don’t have the skills or personnel to manage it.
Right. Best practice isn’t cost effective in such environments, right?
IT as Music Orchestra Metaphor. Yeah. I went there.
So that begs the question, what is hyperconvergence? Scale Computing has a whole article explaining this much-hyped term, but I have a different approach.
Think of the applications we in IT deliver to our users as somewhat analogous to music.
Riker: You’ve got to be kidding me. IT makes music?
Bear with me now.
Now, the traditional way of making music is to get an orchestra together. You hire the very best brass, wind, and string musicians, and each of these highly-trained pros bring their best-in-class instruments into a special, purpose-built chamber that’s designed to deliver music in an optimal way.
Next, you hire a world-class conductor, vExpert or better. It’s the conductor’s job to get all the musicians and their specialized instruments to work together, in harmony, at the right time and in the right sequence. The conductor is uniquely capable of doing this because he’s familiar with all the instruments, and he knows what good music sounds like, and everyone respects his credentials & command of the music.
Riker: Okay. Get to the point.
So my point is that midmarkets, as Scale thinks of them, don’t hire conductors. They don’t buy world-class instruments either, nor do they have professional musicians on staff. What’s more, the entire music industry more or less ignores midmarkets; they think there’s no appreciation or desire for beautiful music in such environments.
When I saw the Scale Computing logo, it called to mind the logo of a midmarket American sports coupe of the 1980s: the Camaro IROC-Z. What was the IROC-Z? Better than a regular Camaro, the IROC-Z gave the working man a taste of the high-end sports car lifestyle, just as Scale brings enterprise capabilities down to the working IT guy in the midmarket space. Deep, right? Ok so that is entirely made up but still, the Logo ….it’s familiar somehow.
But they’re wrong. Midmarket companies still want good music, even if their music is more Led Zeppelin than Mozart’s 5th. And midmarket IT generalists who study & practice their musical skills, want to deliver that great music to their users, they just don’t have the luxury & time to specialize and their instruments suck.
Scale Computing’s hyperconvergence is, in essence, that. It’s an appliance designed to abstract the complexity of compute, storage and network so that it’s easier for midmarket companies to deliver their applications the way big enterprises do, with high-availability, scale out options, VM replication, and DR/backup on the horizon.
Riker: Absurd metaphor, let’s move on. Tell me about the architecture.
Right. So it’s hardware. With Software. Together at last.
All kidding aside, the hardware menu is pretty simple: you can have any chassis you like as long as it’s a rebadged Dell 1U or 2U pizza box, with a big Scale Computing logo on the front.
The chassis, suitably outfitted with Xeons, copious RAM, adequate HDD & SSD, and a few 10 or 1GbE NICs are called HC3 nodes, with the forked KVM HyperCore OS running as hypervisor.
You need three HC3 nodes at minimum to build a proper cluster.
All of this sounds and is quite routine in virtualization with the exception of one thing: the storage is not shared. There’s no filer or block storage array listening at the other end of your vNIC or pNIC, follow?
Scale me baby. Where’s the storage? Oh right. it’s all up and hyper-converged in my nodes!
Riker: No. Not really. Explain it to me like I’m a 3 year old.
Scale calls their storage abstraction system “SCRIBE” which stands for “Scale Computing Reliable Independent Block Engine,” and is, in addition to be a really kick-ass acronym, feels like an awesome mix of some of the best storage paradigms I’ve used.
For instance, SCRIBE abstracts your disks in a way that feels similar to ZFS or Storage Spaces by bundling them into a single logical volume that is presented to you in the management GUI. Only the disks are on different computers, and behind the scenes, it’s mirroring & “wide-striping” that data to at least two other nodes, such that you won’t lose a VM if a node goes down, you’ll just suffer through a restart.
Also, it’s borrowing RAM from each host to accelerate reads & writes of the most recently used blocks across all three nodes. I thought that was cool, but I have similar tech in my Hyper-V CSV caching. But here’s the kicker about SCRIBE: you’ll likely see latency improvements from the VM’s perspective as both disk + RAM may be on the same node’s bus.
The architecture was impressive. There’s a whole paper on it that’s quite in-depth. I’ll say that I like their style.
Riker: Management, Orchestration?
So management is done through an HTML 5 interface called HC3. This was the GUI system they demoed for us at #VFD4, and while it falls short of System Center VMM in terms of orchestration & managing virtual infrastructure, you have to put HC3 in the context of where Scale is trying to sell it: midmarket IT shops that are only partially or not-at-all virtualized yet.
This is your storage. On Scale Computing.
For a place like that? It’s quite game-changing. You can build templated VMs, you can get comfortable with things like snapshots, cloning, and monitoring performance in a converged, shared system. SCRIBE can replicate block volumes off-site, much like Hyper-V’s replication, and it’s all done from within this nice HTML interface.
HC3 supports running the latest and greatest versions of Windows as VMs, with guest integrations between parent/child.
Lastly HC3 is also way better than the vSphere http client, but that’s admittedly not saying much.
Riker: Can you buy the software alone?
I like the software; but no. Scale won’t de-couple software from hardware entirely; they make you buy the whole thing.
In the midmarkets where they play, IT guys have to re-use servers. Wouldn’t it be great if Scale gave such places a non-supported mechanism to run HyperCore on hardware they already owned? Don’t they want more Scale Computing nodes out there, not fewer? Can’t I play with this in my lab?!??
Scale resists that idea and I understand why; this is a highly-engineered stack designed to operate as an appliance.
Riker: So we don’t get root access either?
Doubtful. Which makes me sad, angry and such, but it’s the New Way.
On the positive side, because HyperCore is KVM, the Scale guys told me that virtual appliances like the lovely Kemp Load Balancer will work inside of HyperCore.
Riker: Last thoughts?
Oddly, they don’t sell switching as part of their hyperconverged product. So you have to buy the server hardware, you have to buy the software, and you can’t buy the software apart from the hardware.
But when it comes to the switching stack, perhaps the most critical component in an HA design, you as the customer have to Bring Your Own Switch. Does that make sense to anyone given the market they’re focusing on? Scale tells me they do have recommended switch configs and can resell if necessary, but still.
Word of advice to anyone pondering a Scale Computing infrastructure: make sure you have two good switches. I’d get ‘em with Layer 3 licensing cause collapsing the router onto the switch stack is part of my hyper-converged style, but just make sure they’re decent, line-rate switches and you should be good to go.