SoftIron's Breakthrough in Private Cloud Technology Was On Display at Cloud Field Day 19 [CL92]

[00:00:00] Ned: Tomorrow's a busy day full of family things and so I'd rather soldier on through today and get the last thing that's on my list of stuff for this week done and then I can collapse in a heap in about one hour.

[00:00:16] Chris: I'm rooting for you.

[00:00:18] Ned: I think it's gonna be wonderful and I can definitely still speak good words well in the order right.

[00:00:24] Chris: On purpose, or?

[00:00:34] Ned: Hello, Alleged human, and welcome to the Chaos Lever Podcast. My name is Ned, and I'm definitely not a robot. I'm a real human person with feelings, dreams, throat phlegm, and dizziness. All normal human traits for a person working within normal tolerances. Yes, we are all normal here. Time is not an illusion, and you are not a tree.

[00:00:56] With me is Chris, who is also here. Hi, Chris.

[00:01:01] Chris: How do you know I'm not a tree? Seems awfully anti-arborist of you.

[00:01:06] Ned: Well, if there is one thing I've been accused of constantly, it's my anti-arborist stance. I hold to it. I will cut down those trees if they look at me funny.

[00:01:17] Chris: And the entire state of Maine just turned off the podcast furious.

[00:01:24] Ned: Well, it's their own fault for growing so many elms. I actually don't know if Elm, do Elms grow in Maine?

[00:01:29] Chris: No, that can't possibly be right.

[00:01:32] Ned: I don't know. Poplars? We'll go with Poplars.

[00:01:35] Chris: Oak?

[00:01:36] Ned: Oak is good. Sturdy, from what I hear. Yeah. So, I just got back from a trip to California.

[00:01:46] Chris: Boy, your arms are tired?

[00:01:48] Ned: Everything is tired, Chris.

[00:01:52] See, I always think that it's a good idea to take a red eye because like, I'll reset my clock and I'll only have to deal with like being slightly out of phase with time for a day. But the actual lived experience is not exactly what the plan indicated. I'm tired is what I'm saying, Chris.

[00:02:17] Chris: I got that. Yeah, I got that.

[00:02:19] Ned: [laugh] OK. Also, I was listening to a video and watching a video slash listening to it and the host of the video said, "time is a flat circle," and I was like, that's our joke. You don't get to make that joke. Very disappointed.

[00:02:34] Chris: How dare you, sir and or madam.

[00:02:37] Ned: Yeah, I always assumed that was our thing. I bet they're using AI and it ripped us off.

[00:02:44] Chris: Anyway.

[00:02:45] Ned: Anyway, so the reason I was in California and the topic of today's episode is Cloud Field Day.

[00:02:54] Chris: Whee!

[00:02:54] Ned: Now before you switch off because you're like, "Oh God, they went to another one of these things and it's just going to be like a barrage of vendors." That's not what this is going to be. I instead looked for themes of the presentations of the different vendors and zeroed in on two themes and one vendor in particular that really caught my eye and I would like just explore their solution with you a little bit, Chris, how's that sound?

[00:03:19] Chris: Was the theme Sandwiches? And was the vendor Primos? Because otherwise I'm not interested.

[00:03:27] Ned: Oh, I might lose you there. Day two was a boxed lunch and it was a sandwich.

[00:03:33] Chris: All right. All right. You barely have my attention.

[00:03:36] Ned: Okay. I will endeavor to hold on to that shred of attention. So, l ike I said, last week I attended Cloud Field Day 19 in beautiful Santa Clara, California, where it rained most of the time. And apparently this was my fault somehow. At least that's what I was told.

[00:03:54] Chris: Yeah, that tracks.

[00:03:56] Ned: Cloud Field Day was a two-day in-person event with five different vendors presenting their solutions to a panel of 12 delegates, all of whom were there in person. It's the first time that's happened in a while.

[00:04:10] Now, I'm not going to rehash the whole thing. If you're not familiar with Cloud Field Day and Tech Field Day at large, We'll include a link in the show notes. If you want to watch specific vendor presentations, that link will take you to a page that has all the videos of the vendor presentations. But like I said, I want to focus on themes of the event and pick out one vendor in particular.

[00:04:34] You will not be surprised at all to learn that AI was on everyone's brain.

[00:04:41] Chris: What?

[00:04:42] Ned: I know. Shocking! In this, the year of our lord 2024? Yes. 2024.

[00:04:48] Chris: That sounds right.

[00:04:50] Ned: I really had to think about that for a moment. That's where we're at, folks. What year is it? Who's the president?

[00:04:56] Chris: Things are going well.

[00:04:58] Ned: Yes. So, we heard from a company called Neuroblade, who makes an accelerator card for big data analytics. No relation to Neuralink, Thank God.

[00:05:09] Chris: Or Neuromancer?

[00:05:11] Ned: That was, uh, less clear. So maybe.

[00:05:16] Chris: [laugh] Either way, it's an uncomfortable name. It makes me uncomfortable.

[00:05:20] Ned: Yeah, I see what they were going for. They make an accelerator card that snaps into the PCI bus. And so it's kind of a blade. And the Neuro thing, I guess, because they created their own custom chip that accelerates SQL processing.

[00:05:38] Chris: Okay.

[00:05:40] Ned: Which has something to do with Neuro? I don't, I don't know. Anyway, moving on. The product actually seemed pretty cool. Naming is hard.

[00:05:48] Broadcom presented and talked about their various chips, speeds, and feeds, and especially the Ultra Ethernet Consortium. Which is looking to beat out InfiniBand in the world of AI interconnectedness.

[00:06:03] Chris: But we love InfiniBand.

[00:06:05] Ned: Do we? Do we love the fact that there's one company that makes the InfiniBand and the GPUs? They kind of have a stranglehold.

[00:06:14] Chris: We like the name.

[00:06:16] Ned: What was kind of funny is I asked them to kind of compare and contrast the two standards a little bit. And one of the presenters, or wasn't even a presenter. It was just one of the Broadcom people rather defensively told me that out of the seven major cloud AI companies, six of them are using Ethernet and not InfiniBand.

[00:06:41] He didn't say which one's using InfiniBand, but I know it's Microsoft. And guess who's the biggest, who has the biggest AI clusters in the world?

[00:06:51] Chris: Also Microsoft.

[00:06:52] It's Microsoft, yeah.

[00:06:53] Nice.

[00:06:54] Ned: So, I'm not saying that Ultra Ethernet doesn't make sense, I'm sure it does, but, you know, I don't think that fact did what he thought it was going to do.

[00:07:05] And then Dell also presented, and they are working on building out storage arrays in the cloud to handle - wait for it - AI workloads.

[00:07:16] Chris: Amazing.

[00:07:17] Ned: Amazing.

[00:07:17] Chris: I was not going to be able to guess that.

[00:07:19] Ned: so, a lot of AI. We're not going to go any deeper into that. Because the other big theme was the re-emergence of private cloud from vendors that are not VMware.

[00:07:31] Platform 9, who's been around for a little over 10 years now, their presentation focused on their Elastic Machine Pool product, which is supposed to help you save money on your EKS clusters and eventually other Kubernetes clusters in the cloud. But if you look at their product from a larger perspective, they are very focused on remote management of Kubernetes clusters for their customers. So, sort of managing a private cloud for customers.

[00:07:59] Dell did two presentations, and the one that wasn't about storage was about the management capabilities and integrations that come with their new APEX Cloud Platform, on which you can run OpenShift or Azure Stack HCI. It wasn't clear immediately what the cloud platform was, but essentially, if you buy certain SKUs from Dell, it's a stack of hardware that you then can configure from bare metal to have OpenShift or Azure Static HCI running on it. And it does the full provisioning and then actually integrates with the management console for Azure Static or OpenShift. So you can see all the details about your hardware and you can do patching and firmware upgrades and stuff like that. So trying to make private cloud a little bit easier, more cloudy.

[00:08:50] And finally, the vendor I actually want to focus on is SoftIron. They're a private cloud solution built from the ground up. And I do mean built from the ground up. I'd never heard of this company. And Chris, I don't know if you've ever? No?

[00:09:06] Chris: Sure haven't. No.

[00:09:08] Ned: They've been around for about 12 years, if I'm remembering correctly. At least a decade. And their solution is private cloud. That's what they deliver. And they deliver it in the sense that they designed and built the hardware that it runs on. And I don't mean like, they just farmed it out to an ODM with some specs. I mean, they actually have manufacturing facilities in San Francisco and in Australia to build the system boards and basically the tin that the system boards go in for these 1U rack servers.

[00:09:48] So are they using some commodity components? I mean, of course, they're not making the fans themselves and they'll use AMD chips, but -

[00:09:56] Chris: Right.

[00:09:58] Ned: - a large portion of the actual board itself is all custom built by them.

[00:10:04] They also designed and wrote and built the software that runs the entire stack as well. And when I say the entire stack, I mean, they took Linux and customized it for their own operating system for a level one hypervisor, they wrote their own firmware [laugh] for several of the components that run the system board, including the BMC controller. And they also wrote the software that runs on top of everything for the control plane, the virtual networking, etc. So yeah, this is when I say they built it from the ground up, I truly mean that.

[00:10:44] It is meant to be a true private cloud in all senses of the term. Are you familiar with Oxide Computing?

[00:10:53] Chris: Uh, yeah.

[00:10:54] Ned: Yeah, they've been making waves.

[00:10:56] Chris: But let's pretend that I'm not.

[00:10:58] Ned: Okay, well, they've been making waves recently, and they did a similar thing, but they're a much younger company, but their idea was we will build at the rack level.

[00:11:08] So we will deliver racks to you, and that's the unit by which you order it. And we, likewise, are going to design the hardware, write the software, do everything. So taking sort of a page out of the same book. Now this is not something that JoeSMB is going to go out and buy. The minimum size system that SoftIron delivers is 8U, and that's like a baby system.

[00:11:38] They don't recommend it, but they will actually, they will scale down that far if you're in a situation where you want to have A bunch of these installations in different remote sites, they'll do that. But where they really focus is on service providers or large organizations or especially governmental organizations because they're FedRAMP certified and they have all the other badges on their cloud.

[00:12:03] So governmental institutions can buy their solution, install it in their own data centers, and get what is essentially cloud, but not have to worry about it being hosted on an external cloud if they can't do that for reasons like compliance and other rules. So that is sort of very high level. The reason I think that the time is ripe for private cloud kind of gets back to the discussion that we had very recently about what's happening with VMware. And talking to the SoftIron folks and some other people, they're all actively targeting VMware customers with marketing campaigns basically saying we can get you off of VMware and onto our platform. And they are going to market with that message today.

[00:13:00] Chris: So what you're buying is X amount of these 1U boxes in an 8 being a minimum that they don't want you to really buy. They want you to buy like An enterprise's worth of boxes.

[00:13:17] Ned: Yeah, to give you an idea of like scope and scale, right? What are you actually buying from them?

[00:13:22] You're buying the servers and then there's a cost, an ongoing licensing cost for their software to manage your cloud. And that includes things like support, hardware support. So if something goes wrong with a node in the cloud, they have people who will come in and swap out the hardware for you, or they can ship it to you and you can, you know, swap it out yourself.

[00:13:48] The actual servers themselves come in one of three flavors. There's a networking node, which is where the control plane runs as well. There's a compute node, which, you know, does compute.

[00:14:02] Chris: Computelating.

[00:14:03] Ned: Yeah. And then there's a storage node, which provides software defined storage to the rest of the cluster. So when you, if you purchase that 8u thing, you would get three nodes for the control plane because it's a minimum of three so you can, you know, lose one.

[00:14:20] It doesn't impact anything. So you get three of those interconnect nodes. You get, I believe, two compute nodes and three storage nodes. And so, you actually don't get that much compute out of that because there's all the overhead that you would get on a larger system just scaled down. And that's really a problem with a lot of these private cloud solutions when you try to scale them down to a certain degree.

[00:14:44] There still needs to be somewhere for the overhead to run.

[00:14:48] Chris: Yeah.

[00:14:49] Ned: And, yeah, you need to make room for that.

[00:14:52] Chris: And one of the major reasons for that is, you know, everything has to be wildly redundant for your X amount of nines to actually have a chance of being successful.

[00:15:02] Ned: Exactly. And they wanted to be able to run in a fully disconnected mode, which means this thing doesn't have to dial back to anything to continue to function.

[00:15:13] Chris: You mean back to the SoftIron home base?

[00:15:16] Ned: Right. Unlike something like AWS Outposts, which needs - it doesn't have to be a constant connection, but Outposts does need to be able to check in on a regular basis back to the home base, because the actual control plane for Outposts runs in AWS, not on the hardware running locally.

[00:15:36] Chris: Right. Which is why the way that people think about the AWS Outposts is really you're creating your own secret personal region more than you are creating your own secret personal cloud.

[00:15:48] Ned: Yeah, AWS now has local zones, which from what I can tell is basically just AWS Outposts, but it's managed by AWS instead. [laugh]

[00:15:57] So, that's kind of what you're getting if you get an outpost is you're just you're getting the hardware, but you're renting it You don't own it and it has to call back constantly to AWS. This is not that. This is you own the hardware, you pay us support and licensing and it doesn't have to be constantly connected back. Which some of the government agencies require that it not be dialing back somewhere constantly. So that's what you're buying.

[00:16:27] But, what they were talking about is first of all, the nodes, especially the storage - well, all the nodes in the cluster don't have a boot drive in them. They all boot from network. So if a node goes down or a node goes bad and you need to replace it, it's as simple as pulling the node out, snapping a new one in, recabling it, and then when you boot it up, it just automatically adds itself to the cluster, and you're off to the races.

[00:17:00] Chris: Right, I assume it would be smart enough to know, okay, this is a storage node, it goes into the boot process for storage nodes, and it just tacks on to the end.

[00:17:10] Ned: Yeah, and they did say that in terms of where that node gets placed and who has access to it is something that you can configure, so this is a system that supports multi tenancy.

[00:17:20] So if you're a service provider and you want to offer private cloud services to customers, you can carve their solution up into individual tenants that have separation between each other at either a software or a hardware level. So you can say, you know, these compute nodes and storage nodes are dedicated to this tenant, or I have three tenants on these nodes and they're virtually separated .

[00:17:44] Chris: A lot of it kind of sounds like, and I know we've talked, we've absolutely had this specific conversation and I've made this specific point before, but this really sounds like, you know, the extreme version of just a composable infrastructure.

[00:17:57] Ned: It certainly does.

[00:17:58] Chris: You build up your design and model of what these resources are going to do.

[00:18:02] They are hardware. The hardware is not dependent. It's just. Does it have the right CPUs? Does it have the right storage, etc.? You plug it in, you hit go, when it breaks, you unplug it, you replace it, rinse and repeat.

[00:18:15] Ned: Yeah, and it's very similar to how Azure and AWS handle their hardware in house. Like, AWS servers at this point are designed by AWS. They farm the actual manufacturing out to ODMs, but the actual design and specifications all come from AWS, and they have EC2 nodes that will host compute and then they have storage nodes for like S3 and stuff like that. So if you think about the way that they're building it, it really maps onto the same model that public cloud is using today.

[00:18:54] It's not the exact same APIs, right? You're going to have to learn how to interact with their solution. And that was one of the things that I pressed them on. I was like, Are you just offering basic IaaS services? Or do you have expanded capabilities? Something like a DynamoDB or Azure Functions or something like that.

[00:19:14] They were a little cagey about that one. I know.

[00:19:19] Chris: On the roadmap, perhaps?

[00:19:21] Ned: Uh, what they said is that today they have a marketplace that third party vendors can offer their solutions in. So if you wanted to use F5 virtual firewalls inside of the solution, you can easily do that and select it from the marketplace.

[00:19:37] Or you can curate a marketplace for your tenants, if that's what you're trying to do. So that's certainly an option, but built in, it sounds like, right now, they offer the basics of storage, compute, and networking. And then it's kind of up to you to add any additional features and functionality that you want.

[00:19:58] I'm kind of okay with that, though.

[00:20:00] Chris: Yeah, I mean, it makes sense you can already run Azure functions as software on your own laptop if you want to, and you know, it's not going to run at the scale that it will on AWS, but it'll run fine. So you just do that with some type of a black box inside of this SoftIron private cloud and you're good.

[00:20:21] Ned: And speaking of scale, so we asked them, who's your biggest customer? Like, what's the biggest deployment they have? And they said that their largest cluster today runs at 16 petabytes of capacity. They have a customer in Australia that is doing that. So that's pretty sizable. Um, their next biggest customer is one in America in manufacturing. They couldn't say names, of course.

[00:20:46] That's running a 9 petabyte cluster. And they have several other customers that are in that range. So this thing scales out pretty big. And they said that 16 petabyte customer isn't at the limit of the scalabilities. That's just as much as they need right now. They can continue to add capacity and it's still the same cluster as a whole.

[00:21:09] You can also do things like deploy clusters in different regions and tie them together and do some replication between them. So they do have the concept of both sort of availability zones and regional deployments. That's kind of nice. I like that. The one downside to the fact that this is hardware based and the minimum is an 8U deployment is getting hands on with it is a little challenging.

[00:21:38] I don't think I can just ring up SoftIron and be like, Hey, can you just ship me a half rack for like three weeks and then I'll give it back? That's cool, right? So we had to go by the demonstrations that they showed us. And they did bring in some hardware samples, but they didn't roll a whole half rack of gear into the presentation room to show us what the private cloud looks like.

[00:22:00] One thing I want to come back to is the fact that they wrote their own BMC controller. Because, as you and I well know, the BMC controller is a major attack vector. And the software that tends to run on those things is pretty bad. So, they basically said in order to get the easy FedRAMP approval and everything, it was actually a lot easier for them to say, we wrote our own BMC, we use our own BMC hardware, and none of it is manufactured or written in China.

[00:22:31] And the government was like, "Gold star!"

[00:22:35] Chris: Well, I mean, that does solve some of the problems and concerns about foreign actors and whatnot, but you're always running a huge risk when you start from scratch on a product that has already got a lot of runway behind it.

[00:22:52] Ned: Yeah, I think, ultimately, looking at how poorly most BMC controllers work, starting from scratch is probably the best thing that you could do. Rather than just, like, putting another coat of paint on something that is falling apart.

[00:23:10] Chris: Sure.

[00:23:11] Ned: But it's also a challenge.

[00:23:13] Chris: Is any of the stuff that they're doing open source?

[00:23:18] Ned: I think they're using open source components, certainly. Like, their operating system is based off of Linux, right? But I don't believe that they are publishing any of this as open source solutions.

[00:23:31] So if you're looking to consume it, you got to consume it through them. It's not going to be something you can just do on your own. I would need to check that cause maybe they are publishing their BMC software open source.

[00:23:43] Chris: That's exactly where I was going because the one way you can, a) benefit the community, but b) you get tens of thousands of insanely qualified programmers helping to proactively fix issues, shall we say?

[00:23:58] Ned: [laugh] Yeah. Well, I will say that the BMC software you tend to get today is all closed source and not source available. So at the very least, they can make the source available to some of the governmental agencies if they want to inspect it.

[00:24:15] Chris: Right.

[00:24:16 ] Ned: Something that some of the other vendors are not willing to do. Or can't, depending on the situation.

[00:24:24] So, the SoftIron solution really seemed interesting to me, and I think for people who are looking to move off of VMware, and they want a solution that is not going to require all the layers of complexity that managing vSphere had, this is probably a pretty good option. And you can run it side by side, so you can start with a medium sized cluster of SoftIron, migrate workloads over, and then as you shut down VMware components, you can add more to the SoftIron cluster until you get to a point where everything has been moved over.

[00:25:05] What you do with your old hardware is kind of up to you because SoftIron is supplying the hardware, so it's not like you can take your old servers and install their HyperCloud software on that. It's, uh I don't know. You can depreciate it, donate it to a school, build a really cool sculpture in the front yard.

[00:25:23] I don't know. What do you do with servers?

[00:25:24] Chris: Or a fort.

[00:25:25] Ned: A fort. [laugh] A fort that cuts everybody. Why are servers so sharp? [sigh] Probably because it takes work to dull the edges. If you want them, like, nicely rounded.

[00:25:41] Chris: I don't know. A number of people have told me that I was dull just naturally.

[00:25:45] Ned: Oh, maybe we should make you into a server.

[00:25:48] Chris: Mm. That didn't go well when I was a teenager.

[00:25:50] Ned: [laugh] Well, if you hadn't spilled the whole jug of iced tea on Mrs. Robinson, you would have been fine.

[00:25:58] Chris: [laugh] I think my all time favorite oopsie was when I was helping, uh, it was crazy rush, I was helping bussing tables just because we were shorthanded because, of course.

[00:26:11] Ned: Always.

[00:26:11] Chris: You remember that one restaurant that wasn't shorthanded?

[00:26:11] Ned: [laugh]

[00:26:16] Chris: Anyway, misunderstood how heavy the tray was gonna be and tried to pick it up without like bracing myself and ended up flinging it on the floor, shattering I want to say 30 glasses.

[00:26:26] Ned: Well done.

[00:26:29] Chris: Screeching the entire restaurant to a silent halt. And just feeling the emotion of - Oooh - radiating in three dimensions.

[00:26:41] Ned: Yeah. Did they applaud? That always feels good. That's what they did to me.

[00:26:46] Chris: No applause, just kind of not making direct eye contact and a snicker? There was a number of snickers?

[00:26:54] Ned: Well, I like Snickers.

[00:26:55] Chris: Not this kind. This was like, this was like the fun sized. Like, nobody likes them.

[00:26:59] Ned: Oh, no. This is gonna sound strange, but have you ever noticed that the fun sized Snickers don't taste the same as the full size?

[00:27:08] Chris: Yeah, they taste like disappointment.

[00:27:12] Ned: [laugh] No, I truly believe that like the quality of chocolate that they put into the, into the smaller ones is not as good.

[00:27:18] Chris: Dude, you're getting into one of my crazy food conspiracy theories right now and I am for it. Let's dig in.

[00:27:23] Ned: [laugh] Okay.

[00:27:24] Chris: Do you know what else is not as delicious when it's small?

[00:27:29] Ned: Tell me.

[00:27:30] Chris: Pizza.

[00:27:31] Ned: You mean like a personal pan pizza versus a large?

[00:27:34] Chris: Personal pan pizza doesn't count because I know where you're going you're thinking Pizza Hut and that's different. Set that aside.

[00:27:41] Ned: [laugh] Okay. All right.

[00:27:41] Chris: If you go to just like whatever random pizza shop down the street, Dino's Pizza, whatever, and they have a small, medium, and large.

[00:27:50] You order three pizzas, small, medium, large, exact same toppings? It is a guarantee that the large is the most delicious. Guarantee.

[00:28:00] Ned: Interesting. I feel like I need to test this theory.

[00:28:04] Chris: I think it has to do with the way that the ingredients meld together and the density of the dough. The fact that there's less dough with a smaller pizza.

[00:28:14] Maybe the cheese cooks too fast. Maybe vice versa. Maybe the sauce doesn't maintain the temperature correctly. Whatever the weird alchemy that goes into a large pizza only works with a large pizza. Incidentally, you can get too large of a pizza.

[00:28:34] Ned: Oh.

[00:28:35] Chris: If you ever get one of those, like, uh, what was it, a New York 28 inch Domino's pizza or something?

[00:28:41] Ned: Yeah.

[00:28:42] Chris: Disgusting.

[00:28:43] Ned: Or you should go down to Lorenzo and Sons.

[00:28:46] Chris: Well, and that's why you only get Lorenzo's at three o'clock in the morning.

[00:28:50] Ned: It's the only time it tastes good.

[00:28:51] Chris: Exactly. [laugh]

[00:28:54] Ned: And it's just wildly entertaining to watch other drunk people try to eat that big of a pizza slice.

[00:28:59] Chris: It's a guarantee that somewhere on South Street, you will find almost an entire slice face down.

[00:29:05] Ned: [laugh] All right. Well, that's everything for SoftIron. Went on a little tangent there, but I'm fine with it. Thanks for listening or something. I guess you found it worthwhile enough if you made it all the way to the end through the weird pizza conspiracy. So, [laugh] congratulations to you, friend. You accomplished something today.

[00:29:24] Now you can go sit on the couch, order yourself a small, medium, and large from Dino's Pizza, and let us know if they taste different. You've earned it. You can find more about the show by visiting our LinkedIn page. Just search Chaos Lever or go to our website, ChaosLever.cow, where you'll find show notes, blog posts, and general tomfoolery.

[00:29:44] We'll be back next week to see what fresh hell is upon us. Ta ta for now.

[00:29:56] I really do think that Snickers - Like, there's multiple sizes, because there's the really tiny ones, and then there's the snack size, and then there's regular size and king size, and they all taste different.

[00:30:08] Chris: Yeah, I forget what the, like, the snicker bites. They're like, allegedly supposed to be literally one swallow. It's just like, why?

[00:30:17] Ned: Yeah, you're cheating yourself.