In this episode Enrico Signoretti talks with Leo Leung about Oracle's Cloud, comparing and contrasting with its on-site offerings, getting into details of CPU, memory and network.
Leo Leung is Senior Director of Product & Strategy @OracleIaaS. Product Management, Product Marketing Exec Alum: @Scality @OxygenCloud @EMC. Business and Strategy @TepperCMU.
Enrico Signoretti: Welcome everybody! This is Voices in Data Storage, brought to you by GigaOm. I'm your host Enrico Signoretti, and today we'll talk about cloud computing and cloud storage. My guest for this episode is Leo Leung, Senior Director of product management at Oracle Cloud. Hi Leo, how are you today?
Leo Leung: I'm great Enrico, how are you?
I'm fine. Thank you very much for joining me today. I'm very happy to have you today because we met after a very long time, twice in two weeks. We had this briefing about your object storage option at Oracle Cloud, and I was impressed. And then again, we met at Tech Field Day the week later, and when I got the full picture, that was amazing. I always considered Oracle Cloud just like a side play for Oracle, you know, Oracle legacy applications, the database, enterprise application, you don't think about them as cloud players, even if your company from a marketing point of view spent a lot of money. But I was skeptical. Instead I found some interesting things.
Sure yeah, thanks a lot, Enrico. I think it's natural to be skeptical. I think if you've been in the tech industry for a long time, like we have, there's lots of people that talk about various things that they may be going to market with or new products or new features. I think it's very natural to be skeptical, but I do think that if you take a look at the cloud market, I very much think we're still at the beginning. Much like the general enterprise on premises market, this is a market that's hundreds of billions of dollars, and we're just at the beginning, even if you consider that we're maybe 12 years in from the inception of AWS. And our belief very much was that there was a really big opportunity to serve a very underserved market, which is enterprises -- and in particular their mission critical applications. And we felt like we had a unique opportunity, as Oracle, to serve those customers. I'll talk about it more later, but I think this wouldn't be possible if you were thinking about Oracle as this monolithic entity.
In reality we've brought in literally thousands of experienced cloud engineers, architects, product managers, and leaders into Oracle. Many of us are brand new to the company, and bringing this kind of experience as well as start-up experience, in addition to the large amount of investment Oracle has made is really making this possible. So back to your original comment, it's very natural to be skeptical. I was skeptical myself. But if some of those things were not true - the large amount of investment, the true technology and design from scratch, and then the development of such technologies over the last four years, we could just be talking… like some of the other folks in the market.
I have to say that the first time when we spoke, it was all about storage, and you impressed me when you mentioned SLAs or very high-performance instances and it's unusual. Without even comparing the prices which, at least on paper for what I [can] see, they are quite impressive. But the fact that you give an SLA… usually when you go to the cloud, you say, oh, this is a commodity resource. If you don't have enough, you buy more, but yes, you can't expect too much of a consistency. So that's all to be managed from the application and so on.
But actually for example, when we saw that demo with JD Edwards, there is a video which is quite interesting because you demonstrated that you were able to bring a traditional application, legacy application, like JD Edwards – an ERP on the cloud, but you had all the architecture, from 10 application servers, the backend database, and it worked very well. The guy on stage demonstrated fail overs and stuff, and this is something that was very well done from my point of view.
Yeah, thank you. We've been to a number of these kinds of demos and bakeoffs, and I think we always surprise people because we'll pick a hard app -- and not to say that there's anything wrong with doing a machine learning demo or doing some other natural language processing or whatnot -- but this goes back to our focus.
We want to make customers comfortable that we actually understand the kinds of applications they have now. Not the ones they want to build, not the ones that are in the future. We care about those too, but when you think about enterprises, I've heard this statistic cited a number of times and I believe it: most enterprises have on average about a thousand or more applications in their portfolio. And our belief is probably about 80% of them are traditional applications, things that were potentially built even 20 years ago, where the architecture is highly dependent on a relational database.
The expectation is all the things they expect on-premises, and you and I having a storage background, enterprises expect, that's why they've paid for enterprise class SANs, that's why they've paid for enterprise class converged infrastructure. They have a certain expectation around performance, around reliability, around a feature set, that's going to allow them to support these more traditional applications. And we wanted to create an environment that was focused on that --not net new, brand new application where frankly you have a lot of control over your architecture; you have a lot of control over how you sort of spread your risk.
If you're dealing with an existing application and a packaged application, there is a set of best practices that exist, that allow you to run, operate, and protect that application that you actually don't want to change. So, forcing people to absorb the effort and cost associated with re-platforming, for again, like 80% of a thousand applications, so let's say 800 applications, doesn't really make a lot of business sense for most enterprises. So our belief was we're going to go and create an enterprise-like cloud environment. And we'll talk a little bit more about how that's actually very applicable for new applications as well.
We're going to go do that, and we're gonna do things like essentially create a SAN in the cloud. We're going to create an enterprise class file storage service in the cloud because your example about JD Edwards and pretty much any other entire enterprise web application, is they pretty much all expect a certain amount of block storage, a certain amount of file storage, and a certain amount of database centric storage, whether it's local or network attached. And we wanted to give all those things to the customer.
And, in fact your primary strategy is to go after your customers, okay, so demonstrate to them that you are the perfect cloud for guys that already have Oracle on their premises and other Oracle applications in general, but also theoretically at least, you could say that it doesn't change much at the end of the day. Right? So any sort of enterprise application…
Yeah, definitely. I think you've done this analysis too, if you look generationally about what the applications were in particular generations. It's absolutely true regardless of whether it's an Oracle application or if it uses Oracle database, there's sort of a general architecture of the last, let's say 20 years where the expectation is pretty similar from an infrastructure perspective. So, we have over 430,000 Oracle database on premises customers and some of those Oracle applications, some of those are third party applications like SAP or custom applications, that may or may not be developed on again, at more traditional middleware stack. Our belief is we can move all those applications into an environment where they can move it with zero re-architecture, they can move it with some re-architecture, for example, taking advantage of cloud-based object storage or other types of cloud-based services that will automate some things or they can re-platform more completely. We have some other customers that are starting to package their, let's say application or presentation layers in containers, which then makes the whole application stack more lightweight, but they can choose any of these things on their timescale, not even to get to square one.
And the SLA thing you mentioned, again, is part of this focus on, we're going to give you an environment that you can trust that potentially even gives you better SLAs than you have on premises to yourend users, where the guarantees are not just around availability, which means you can reach a server, but there's no guarantee at all about whether the server is going to be performing at one percent or a hundred percent. All the other cloud providers basically wipe their hands after that. If you can reach the server [they say] their job is done. We believe our job is notdone. If you're going to serve an enterprise application, the storage has to be working. The network interconnect between the various tiers of the application need to be working. And yes, of course you have to be able to access the application, but all those things have to be true in order for it to be an enterprise...
And you didn't mention the bare metal instances that you have, so you can give the standard way of implementing applications in the cloud, but also you get a physical server to your customers somehow.
Yeah, we have a range of options, so customers can select a VM based compute like they can in other clouds. They have an option to have only network attached storage or local storage. We give them most local, very low latency, fast NVMe SSD for those VMs, we have GPUs that are based on VMs. So, there's all these VM based options, and yes, when we first came on the market in 2016, the initial offering was bare metal, and now we have multiple classes of bare metal including CPU based, GPU based, as well as a brand-new instance type which is based on the AMD Epic CPU. So with the bare metal, again, there's an option of either having a ton of local storage up to 51 terabytes of AMD SSD locally or again, just network attached or, of course if you have local storage, you can also attach network storage. And the big thing...
Say it again. 51 terabytes?
51 terabytes, yep. That's the biggest by far. Yeah. So what you can do with that one big use case is yeah, you can actually run relational databases on that infrastructure, and that's pretty big even with triple mirroring, right? You're still talking about over 10 terabytes of usable database space. And now we're talking about, we've tested this publicly multiple times: we're talking about 5 million plus IOPS, which is actually considerably faster than what most people even have on prem.
Exactly. So you're talking about instances that can lead to not just enterprise use cases, but you know, HPC, big data, media entertainment rendering, kind of workloads, not just enterprise.
Yep. These are very beefy instances. You can spin them up in a few minutes, or you can basically Terraform it or script it and you can spin up 100 instances if you want in minutes. And yes, people can run very, very intensive workloads, compute intensive, memory intensive, storage intensive types of workloads on these machines and we have many customers in, as you say, the classic HPC space, in the rendering space. Last week we had two big testimonials. One was Altair, who has built software as a service (SaaS) on top of our infrastructure that's all about computational fluid dynamics. These are super rigorous, parallel, high performance types of applications running on us.
The other really cool testimonial -- again, this is all public -- is from Cisco. Cisco is increasingly becoming a software and security company. They launched their security software as a service, which they called Tetration. This is a large-scale sort of log analytics platform and they're using those bare metal instances and they were able to see...I'll send you the video… 60 X, 60 times more performance than other clouds they used because they were able to implement their proprietary software on top of our bare metal instances, clusters of those instances.
You're talking a lot about CPU and memory and...you didn't mention networking. And again, networking is quite complicated usually, especially not for primary service providers. Connectivity is important and not just in the data center, but how you provide connectivity, direct connection to your customers. What can you say about this?
Sure. Again absolutely true. The network is critical whether you're talking about network connectivity into the cloud, connectivity between your own environment and the cloud as well as the network inside of the cloud. And from the very beginning, we invested heavily in terms of having both a very high-performance cloud network as well as making the choice to not oversubscribe it. The way I explain it, usually to people that are not familiar with the cloud is most clouds are very much like Southwest Airlines or Orion Air. They will sell more tickets than they have seats for. And at rush hour you may get bumped, right? You may encounter congestion even getting into your seat. And that's a choice. They don't have to sell more tickets than there are seats, but they do because that's their business model. And that is very much the business model of most providers.
We made a choice to do two things. Number one is design a network that is hyperscale, very high performance, 25 gig network, but on top of that we said we're not going to oversubscribe. Every tenant will have full access to that network, node to node. There will be no noisy neighbors. So that was a big choice that we made, again, going after a very specific audience and that's why a customer like Cisco can see that kind of performance advantage. On the connectivity side, again we've chosen to number one, offer very, very low-cost bandwidth where you don't get penalized heavily for taking data out of the cloud.
Typically, the models that exist in the cloud are the outbound bandwidth is quite expensive and it can become a very large part of your bill. We offer 10 terabytes for free every month and then a very, very low price after that. So very low cost outbound, which means you can actually run your applications without penalty, and maybe even more importantly if you're an enterprise, you probably want some kind of direct connectivity into the cloud, which means a dedicated connection which we call Fast Connect. And again, we offer that at extremely competitive pricing where it's not actually a bandwidth charge, it's just a port charge. So, you can buy a 1 Gig port, a 10 Gig port or some combination of those things, and you get to use that as much as you want and you're only paying a fixed price per month. And when you look at a combination of those things, we're offering extremely low pricing when it comes to access into the cloud, extremely high performance in the actual cloud, SLAs around the network connectivity, SLAs around the connection into the cloud. We think it's a very competitive offering.
I think that every time you talk SLAs somebody in the enterprise will listen to you because this is not the usual conversation you have with cloud providers.
Yeah. And I think we like that conversation. Again, a lot of this goes back to what you said in the beginning. I think that people think this market is already done. And it's true, it's north of $30 billion now, but we think it's just beginning and really, we just want an opportunity to have a discussion. And I think there's a lot of beliefs about the cloud infrastructure market that are actually no longer true. Right? There has been some bars set by some of the original players in the market that have been broken at this point. So all we're looking for is an opportunity to have a conversation.
As you said, being that late to the game is an advantage because you have the opportunity to look around and check what the others did and then design your infrastructure taking into account mistakes made by others or new technology that was not available 10 years ago. It's a hub for these hyper scalers to change quickly the way they design at the centers, they deploy their racks and so on. So maybe being late, it's not that bad at the end.
Yeah. I think that's been proven over and over again in technology, where timing is important, taking advantage of new capabilities is very important, and then focus. I think that we absolutely have been able to take advantage of the fact that NVMe SSD is the new standard, right? We use that everywhere, in our infrastructure with the only exception being object storage. Our network is completely software defined. There is no specific networking appliances in there and… 10 years ago SDN was very immature, right? You would have had to write almost everything yourself. We have invested there and again, that's what gives us the ability to provision entire servers without any of our code on the servers. That's what allows us to provision them in minutes. That's what allows us to provision entire Exadata database appliances in minutes.
Taking advantage of those newer technologies is something we definitely have focused on and every time we roll out a service, we're trying to see how we can change the game. And I'll give you two more examples. One as I said, we're the first provider that is offering a compute instance based on AMD, in particular bare metal instance plus VM instances, and because of the economics of AMD, because of the economics that we're able to leverage, we are now able to offer instances for three cents per OCPU per hour. That actually means one and a half cents per VCPU, which is sort of the current standard in the market and that's at least 60 to almost 70 percent less than the lowest cost instance of any other provider. To me, that's breaking a bar. The second thing we're doing is we're going to be formally announcing the GA of what we're calling ‘clustered networking.’ This is something that, if you look on premises, happens a lot with NPI types of applications in HPC where they have massive parallel types of applications where each of the instances need to talk to each other at low microsecond type of latency.
It turns out that databases also benefit from this type of architecture, so we're going to be rolling that out very shortly where customers can now programmatically create a clustered network, initially with what we're introducing, our HPC types of instances -- so very high clock speed CPU instances -- but going forward our strategy's going to be: you can build a cluster out of anything. You can build a cluster out of regular machines, you could build a cluster out of database-oriented machines and all of these are running on the same backplane, which means you can start to push data at this 100-gig standard across this cluster. An example would be if you're running a data warehouse type of database service and you want to take some of that data and actually run it against some machine learning algorithms and our GPU instances, you will be able to tie these together in a cluster at that 100 Gig mark. So really improving the experience around time to results for all different kinds of workloads. So we're super excited and I think again, these two things would be very difficult for an older provider to launch so quickly. We have the advantage of being able to...
Yeah, in fact, you are talking about next generation networking, 100 gigabit, potentially rocky. If you didn't plan for that, now it's really easy to start applying this kind of technology to all these instances, all the servers and even building now, you have to maintain some sort of compatibility with the older instances, so it's quite difficult.
Yeah, absolutely. If you look at large scale clouds, we're constantly going to be innovating, adding new capabilities, adding new technologies and you're absolutely right. Once you architect a certain way and you have a certain amount of critical mass, it's much more difficult to adopt new technologies.
One of the most important aspects now for especially distributed enterprises is: ‘Where are your services available?’ And I'm sure that you are available in the US to at least a few zones there. But what about the rest of the world?
We're currently GA in four cloud regions. So there's a couple in North America, one in Phoenix and one in Ashburn, and there's two in EMEA, one in London and one in Frankfurt. We recently announced a very aggressive plan to essentially add more than 12 regions over the next year. So almost one a month. The first ones are going to be coming in the next few months, first in Toronto and then US government regions, and then we're going to have an aggressive expansion into eastern Asia, so in Japan, in Korea, in India. And then subsequent to that, additional expansion into Asia as well as EMEA. We know that there's a couple of things to think about when you think about regions. One, I do think that people are perhaps a little sensitive to having things close to them. In reality there's a lot of workloads that can really run anywhere, right? It really depends on which customer base you're trying to serve, not where your company is located.
Yeah. But you're telling this to any European… [and there’s] GDPR. Everybody wants database close to his country.
That's fair. Yeah, I think that's absolutely true. When you get beyond just latency, if you're just concerned about latency, there's lots of ways to solve that. We have extensive relationships with all the network providers as well as the exchange networks. So you can get pretty good latency no matter where you are really. So, I just kind of wanted to point that out, but absolutely. If you're thinking about data residency and you're thinking about various compliance laws, then yes, we do have to be in many, many places and again, I think we have a very aggressive plan to be in all the places where people need us to be locally. I think there's huge amounts of demand coming out of Japan, coming out of India. So, we absolutely have to serve those markets. There's also additional demand coming out of South America and other parts of Europe. So, we're absolutely cognizant of that.
The story about Oracle Cloud is really interesting at the end of the day. And again, I wasskeptical but mostly it was my ignorance about the real quality of the cloud and some of it also came from the general idea that I have about the cloud. And I'm sorry to say that because Oracle is one of the biggest software houses in the word, but actually the fact that you guys are quite a monopolist when it comes to database and you're seen as a legacy company. So sometimes you don't think about you as a player in the cloud. I know now that I was wrong and I'm happy to record this episode today just to show and to let you present this different idea of cloud. But, do you also have a way to test the Oracle Cloud? Is there any way to test the Oracle Cloud?
Yeah, absolutely. Again, I think we do have a lot of work to do. I think we all recognize that. I think you've put together some of the heritage of Oracle, which good or bad, I think we do understand enterprise used cases really well. We understand business, enterprise business requirements really well. And we've put it together with a very large team, several thousand of us that have a lot of cloud experience, have a lot of new application experience and, again, humbly I think we do have a lot of work to do and we have to prove a lot. So thanks for the opportunity.
I think that when it comes to free trials, we have a free trial. It's essentially $300 worth of credits, but in reality, it'll last more than that. And you can just go to cloud.oracle.com/try it and sign up for a free trial. It's that simple. And one of the things you're going to see is, there are actually a lot of services beyond infrastructure, and going forward I think another thing that we're going to be able to offer which other cloud providers will not be able to offer is a very, very large portfolio of software as a service, all under the same roof.
We absolutely embrace open technologies, open source, the ecosystem. Again, I think we have a lot to prove there, that we are not the Oracle of old necessarily, so we absolutely accept and embrace those types of applications that people want to bring into our cloud, but we do believe we're going to offer a very large integrated stack of things as well for customers that want that sort of experience. And again, right now you can try our infrastructure in a number of different platform services using that URL I shared, and we can share it of course when this gets posted. And feel free to reach out. I'm very active on Twitter and LinkedIn and always happy to talk.
What is your Twitter account then?
It's lleung. So, my name.
Fantastic. Leo, thanks again for your time today. It was a very nice discussion and again, I suggest everybody to take a look at Oracle Cloud at least, there is $300 worth of free services, so give it a try and I think Leo did a great job today to describe what they can do. Yes. The ecosystem is not as extensive as for other clouds and maybe there are other things that we didn't cover today, but I think that the more competition, the better, right?
And again, thank you Leo. And bye-bye.
Thanks so much Enrico.