Last week I had the pleasure of catching up with GE Cloud Infrastructure Architect Cody Hill when he swung by our Silicon Valley office. In a Q&A session with the Platform9 team, Cody shared his perspective on cloud computing, public vs. private cloud, OpenStack, containers, SaaS and more. Here are the highlights:
Sirish Raghuram (SR): Cody, we met at a financial analyst dinner at VMworld this year. When you stood up and said “I work with GE Healthcare on the private cloud” the whole room stopped to listen.
Cody Hill (CH): Yes, GE is a big company and everyone is excited about what we’re up to.
SR: So what is GE trying to do with cloud computing?
CH: GE’s initial goal was to get to the private cloud. We standardized on vCloud Director, which was good initially. But VMware told us we’d have to move from vCloud Director (which they were sunsetting) to vCloud Automation Center (VCAC), and we weren’t impressed with that product. It was a challenge to setup and manage, and as we were working through it we asked VMware to host it as a service to make it easier for us to consume it inside our datacenter. They said it wasn’t the direction they were taking. Since we were going to have to retool and do all that work, we decided that we didn’t want to go to another proprietary tool with proprietary APIs. So we wanted an open standard, hence OpenStack.
We then started on the journey to find an OpenStack vendor. We started with VMware Integrated OpenStack (VIO) about 9 months ago. And again… It was a challenge to setup and manage. When we heard what Platform9 was doing, we were like “Yes! You guys get it!” and wanted to hear more.
SR: How big is your private cloud today?
CH: We have 10,000 VMs spread across 3 continents with deployments in France, Beijing, and Milwaukee.
SR: Who is using it? Do you have a lot of developers doing self-service automation?
CH: Everyone from business users to developers. If you wanted to host your meeting minutes, you could spin up a wordpress site to do that. Or if you need to deploy a complete Tomcat stack or Weblogic stack with an Oracle database, you can do that. Infrastructure-as-a Service (IaaS) or Platform-as-a-Service (PaaS) – we have all the products.
SR: You spoke about how you’re fairly large scale, probably among the biggest vCloud deployments in the world, and that the end of life of vCloud Director spurred you to try OpenStack in the first place. You had this desire to have a product that looks a lot like Platform9. Now that you’ve worked a bit with Platform9, what’s the experience like so far?
CH: Getting Platform9 deployed day one was shockingly fast. We got it up and running in our dev environment within an hour, spun up a VM very easily. It was probably the quickest deployment I’ve ever had of any product. You can’t even install ESXi that quick.
Then there’s all these things we can leverage. Literally, if I wanted to, the only thing I would have to do is port my templates over to Platform9, copy my some of my automation scripts into the interface and we’re done. Then I can consume it via the OpenStack API, which is amazing, and start deploying VMs. I was floored.
And we really like that you layer on top of vSphere and don’t interfere with the day-to-day operations to manage the vSphere stack. You give us all the benefits of OpenStack without hindering us – no other OpenStack product gives you that.
SR: When we were designing the product, we had a lot of back and forth about whether we should assume a greenfield environment that’s being built from scratch, because engineering-wise that’s a lot simpler than discovering existing environments with existing resources and being able to reconcile the operational changes that are being made, even those made through vSphere directly. What’s your perspective, having operated vCloud Director at that scale?
CH: vCloud Director, VMware Integrated OpenStack – for that matter any OpenStack product that I’ve messed with on top of vSphere – assumes that you will not make any changes in vCenter. You cannot power off a VM, you cannot migrate a VM; you cannot do anything without the knowledge of the control plane. That forces you to completely retrain your entire operations team on how to manage and troubleshoot a system. That’s hundreds of employees who we have to teach “Now this is how you troubleshoot a system through vCloud Director” and “Now this is how you troubleshoot through OpenStack.” We fought that battle, and we had a lot of messed up systems, a lot of restoring from backup, a lot of VMware support cases that said “I don’t know how this got messed up and I don’t know how to get it rewired again.”
You guys building your product from the beginning to be able to make the changes from both directions [at the control plane and directly through vSphere] is amazing. Because we can tell the ops guys to simply use the same tools they’ve already used for their entire vSphere environment. Our private cloud is 10,000 VMs, but our overall environments are probably closer to about 50,000 VMs. And to just deploy this layer on top of all that vCenter and deploy to any pocket of vSphere anywhere and manage it the same way, that’s huge!
SR: What else do you see happening in the cloud space? Is KVM an emerging alternative technology? What about Docker? Or is VMware just the standard? I’m curious about GE, but also more broadly in the infrastructure community.
CH: Everyone would agree VMware has best of breed hypervisor, it’s the best in the industry. But it comes at a cost. It’s not free, it’s expensive. Where is KVM better than VMware? Cost. Take cost out of the picture like GE has done with VMware [through very deep discounts], KVM really can’t beat it.
Docker, on the other hand, is a technology that’s exciting. As the cloud infrastructure guy at GE, I think we need to provide people the ability to provision containers. But being a big company, and having legacy development teams, we’re wondering if we do implement Docker and get it up and running, is anyone going to use it? Or is it just going to be “If we build it, they will come.”
SR: Have you had people ask you for Docker support?
CH: Not one person. The only people saying “Hey we should get Docker in here” are on the internal cloud team.
SR: What about SaaS? Platform9 is a SaaS product. How open is the world to SaaS-based management fabric for managing enterprise data centers?
CH: SaaS is growing like wildfire, especially within GE. We’ve moved to Service Now, we’re moving to Box. We’ve moved to Cisco Webex Connect, and we’re gearing up for an Office 365 migration. SaaS means one less thing we have to monitor and maintain. CMDB, DNS, IPAM, LDAP integrations – we have to do all this integration and automation once an IaaS system is spun up, and then we also have to keep track of the cloud control plane. So, if we can let you guys worry about the control plane – all the upkeep, all the upgrades – it allows us to focus on the rest.
SR: How important is a product like Platform9 for a company like GE?
CH: What you guys are doing here is probably one of the most innovative things I’ve seen recently. All the problems that you solve – who wouldn’t want all the benefits of OpenStack without the headache of managing it? I can just consume it, and I can focus my time on integrating rather than fighting with a control plane. If you’re like GE and you have to deal with a ton of regulations where you have to provision VMs inside your four walls because of the FDA or HIPAA compliance or whatever it is. Why wouldn’t you want a control plane in the cloud that you don’t have to manage? And who else provides that?
SR: Doesn’t VMware Integrated OpenStack do that for you?
CH: No. I have to manage it. You need a team of 15 people, according to VMware, to manage VMware Integrated OpenStack, at a minimum. That number includes all of the infrastructure people you’d need just for vanilla vSphere as well, but we’d need to add 10 to our headcount just for the control plane.
SR: Have you tried any other OpenStack providers?
CH: We tried Mirantis. What I like about Mirantis is that they didn’t have an agenda. As professional services for OpenStack, they don’t care if you use KVM, VMware, or whatever. However, it wasn’t just a simple click to deploy thing. It’s heavy-duty professional services.
SR: Backing up a little, what are the specific benefits to OpenStack that you value the most?
CH: The open API. We spent two years developing a really tight, integrated systems on top of VMware vCloud Director. Now that they’re ripping that out from underneath us, what do we do with all that code we wrote to integrate and automate? It’s gone. It’s completely gone. OpenStack’s promise is that they’re never going to fully sunset the APIs that are available today. So if we decide to build to the Keystone v3 API, that API will still be there in five years. We don’t have to retool. We can and we should if we want the benefits and features of later versions of the API, but we don’t have to. That was the biggest thing. Let’s standardize on OpenStack so we’re not forced to refactor every couple of years.