Controlling Cloud Costs - The Hybrid Option

Auto-generated captions from YouTube, not hand-corrected, so names and technical terms may be imperfect. The video is authoritative.

all right I don't want to keep you it's been a long day I really appreciate you coming to the CTO advisor hybrid cloud virtual event thanks to kidar all the speakers vfares for helping us put on such an amazing event we're going to talk cost optimization into three different approaches to cost optimization our speaker David Linthicum from an earlier session shared that it's responsibility is to return value to the business there's two ways in which it can return values one via capability we can accelerate the

business whether we're talking about building an Erp system that enables better customer experience or digitally transforming an organization to work with are vendor ecosystems or optimization of cost of these systems and these one or two ways we're returning value to the business today we're going to talk about cost optimization and the three different options as I see them in the industry return value to the business by Saving cost now you may be arguing Keith just do all cloud cloud is the present and the future to

infrastructure and I will return back with Cloud as a operating model not a place the whole point of this conference is to talk about adopting cloud-like capabilities without sacrificing cost and Agility now on paper that sounds about right but in practice there's challenges I'll refer back to some content that we did with a bunch of folks over the past few months specifically David lithicum where we talked about repatriation and the challenges and opportunities with repatriation we'll hit on a little bit of that today but I

highly recommend you go back go back and listen to that podcast via the website https colon 454 slash the ctoadvisor.com I'm an engineer sorry for the uh prefix but no matter where created technology it will always age it will always become Legacy technology you will always incur technical debt and there will be more efficient ways to deploy and Implement Technologies no matter what your architecture is so we're going to talk through these three different Visions one is optimizing in place wherever your cloud is at if

you're in the public Cloud optimizing in place in the public cloud repatriation or do you just take a hybrid Cloud approach throughout your journey let's first talk about the argument for optimizing in place where you're in the public Cloud we'll do this via examples stories I've heard from the field and people taking their traditional applications their traditional app Cloud apps and modernizing them in place to optimize for cost the first example I get is from SATA interview the CTO of SATA miles Ward on the podcast

and one of the stories he shared is when he was working for the Obama campaign and he built out this massive infrastructure that had a bunch of load balancers and the cost for load balancers themselves were in the tens of thousands if he was to build that application today he simply used Route 53 and AWS which has that functionality built in and he would save the tens of thousands of dollars uh from building load balancers the let the key lesson is in this while the price

of individual cloud services may not be going down Cloud vendors are adding more and more capabilities to each one of their services and as a result it lightens the load on the customer side to rebuild some of these non-differentiating features ten years ago I would not have said load balancers starting differentiating today load balancers are non-differentiating so Point number one take advantage of the advancements in the cloud infrastructure Point number two is to look at your patterns I talked to a data protection company doing AWS

reinvent and their representatives have shared with me how they looked at the data patterns of their uh product and their customers there was like a hundred to one they back up a hundred times before they need to do a single restore so if you look at the that pattern of their data flows they're typically ingesting data versus recovering data and they adjusted the tiers of storage instead of having their storage sitting on S3 they migrate data after a day or two which is when it's most

likely for a restore to occur back to tape based backup in AWS so Glacier storage they're using Glacier storage you might think oh Keith Glacier storage is so expensive to restore the data protection company using glaciers seems like the most inefficient thing to do well if you have this pattern where you're only getting one or two restore requests requests a day and most of that is for data that has been deleted within the past day or two this pattern of putting data in Glacier becomes super

economical and you can uh in according to this data protection company reduce your storage costs by two-thirds or more an amazing cost savings without needing to repatriate all of that data to a co-location pay as you go versus reservations again optimizing for cost and place looking at commitments and committing pre-committing your spin to a cloud provider if you prepay you will get breaks so that's Point number three p prepaid for commitment so you see all of these articles talking about uh companies committing a half a

million dollars a billion dollars to the cloud providers this is their way of hedging costs we did it in the data center by buying excess capacity of servers up front and then we could put uh we could control demand based on our capacity this is no difference you're prepaying for the amount of capacity you believe you can control for not that you'll use but that you can control for prepaid versus pay as you go the fourth area is to understand your slas do you really need

a always on VM SLA for a VM instance that doesn't have that service level agreement akaa web server for a intranet if the intranet web server can go down why would I pay dollars a month or 300 or 500 a month to have that VM always available when I can use spot instances and if the VM goes down now we start another one I can create an AWS Auto scale rule that says that there needs to be at least one of these VMS running and every

time it fails it restarts it into another spot instance easy way to save up to 90 percent a cost on a single VM instance then we can get into re-architecting the platform these other examples were cost saving hacks the other option is to just completely re-platform the application rebuild the application to be container uh or microservices Centric moving it to VMS to containers and microservices or even starting all over and building the application based on serverless Technologies we had a really great podcast with this then

CTO of New York Times that talk through that motion of adopting serverless Technologies talk about the other argument the argument for repatriation first let's define repatriation there's some confusion in the market I'm saying anything that's not public cloud is repatriating your workload if I'm not running is in AWS Azure gcp oci IBM we're talking about repatriation so this is equally it's qts tier Point all these co-location providers I I could be consuming bare metal in their data center I could be hosting racks in there or

I can go as far as building a data center Sticks and Bricks power and Cooling and managing the the physical real estate again all of that falls into the category of repatriation as I'm defining it so let's talk about the workloads that are not great for cloud non-stop applications one of my favorite to pick on is sap I can't really stop and start ECC it is going to be a monolithic application that runs all the time when I scale it I scale it by adding more

VMS not more containers that I can uh add and remove but adding more VMS that process is not seamless I can't simply put a load balancer or a rule all those scaling rule in front of ECC which is the the main module for sap and say at more ECC app nodes once I add them I very rarely subtract them so so this is a very heavy monolithic application that scales for the greatest domain uh the greatest demand for the app that's just not a cloud friendly

application while there's applications or subsets of the application or environments I can use in public Cloud such as Dev and QA which I can double up bring down bring up that might be appealing in the public Cloud but production is rarely appealing from a cost perspective in the public Cloud I bring those workloads back on-prem some of these Mission critical apps alone could justify the cost of moving into a Colo and providing connectivity up to the public cloud a note about latency when we move those

applications from the public cloud back on-prem or on-prem to the public cloud latency census of applications what we call both ones typically have to come with those applications as well predictable demand we get to a point where we know the app no matter the architecture every December we're going to scale up to a thousand nodes and for the rest of the year we're at uh we're at 500 nodes there's some debate here you could build for the high demand or you can build your overall data

center uh resources to flex up to that demand or you may determine that the cloud is the best model for it we'll get into this option in the hybrid Cloud again but there's this idea that if you have the capacity in your private Data Center and you understand a workload's demand you don't need the elasticity of the public Cloud it's probably best to move that that workload back to the cloud remember back to the private Data Center what happens when you have a mature API let's

talk about the process to the maturing API there becomes a point where you don't need the innovation of the cloud for its Civic business process the development of applications just has just settled there you're you don't have developers actively adding new features you mainly move into bug fix and security updates and you don't need the latest and greatest apis this is stuff that we can easily recreate in the data center using kubernetes my favorite buzzword kubernetes to recreate that interface into the data center I think

we're at a point to say that it's comfortable that VMware uh openshift all the major vendors have provided the basic building blocks for you to move cloud like workloads back in the data center that don't need elasticity what happens when we have known Network patterns or network patterns that are so expensive that we need to consider egress if you're at the point where you need to consider egress sd-wan has moved so far to the future when it comes to controlling latency and relying on the public

internet for primary connections that you can have tremendous savings by re-architecting your applications to put these apps into the private Data Center and have them ride over sdn to save tens of thousand dollars a month on egress again something that easily covers the cost of a couple of iraqs into a Colo location in a few 10 gig internet links all right last option and best option all roads lead to hybrid it's the premise of this conference I don't care if you're all in the public cloud

or if you're all in the data center at the end of the day you're going to find that your environment stretches between private cloud-like capability this repatriation slide that we had earlier up and public Cloud agility you're going to need a mix of both most organizations don't just simply cut off Innovation they still need to innovate the public cloud is the best place to innovate the private data center or the private data center model is the best place to optimize for cost what does this look

like practically let's look at even the all public cloud model I did not understand and I have to give VMware and AWS IBM and provide his credit I did not understand the model of having them work cloud in AWS why would I rent bare metal servers to lease vsphere capability well if I want to want to Outsource that low level work but I want the economics of scale of VMS and over subscribing my infrastructure AWS vmc on AWS is a Best of Both Worlds type of

mixture yeah I'm going to spend close to 200 000 a year but if I take that sap application and put it onto a vsphere cluster in in my public Cloud whether it's AWS Azure gcp oci I'm going to have some pretty appealing economies of scale when I put this back on a environment that I can over subscribe and I'm familiar with managing now there may be some questions around licensing you can deal with sap and your software providers about that but it's undeniable that when I'm

putting vdi workloads uh these workloads that always need to be on onto these types of hypervisors that are bare metal in public Cloud I get the advantage of being in the public cloud and I also have the advantage of someone else managing the rack stack and care and feeding of the hypervisor and bare metal layer the other model is that more traditional model where I have private Data Center whether it's Colo or own assets connected up to the public cloud and this idea of cloud bursting

again I am finally at a point where I'm comfortable to say that kubernetes and the architectural infrastructure is there that cloud bursting is real I'm seeing this in the field specifically Walmart shared how they're using their triplet platform in-house platform for managing multiple Cloud instances to handle Cloud bursting a lot of it is manual let's not get it too far ahead of ourselves this isn't magical uh kubernetes picking the best place to run something a lot of this to Walmart's admission is still manual but when

you're talking about cost optimization you can get some serious advantages for having Colo and a couple of racks running on some kubernetes clusters some orchestration on top of that that handles the management of infrastructure between public and private Cloud David Linthicum talked about this during his presentation earlier today my ending point in this short message is that hybrid cloud is where we're going to end up whether you try to or not five years from now that kubernetes based applications that you look built today will be

looked upon as Legacy as a hybrid motion it may not be simply about VMS but these concerns around Network costs storage optimization higher level services around load balancing these types of of optimizations are going to continue to present themselves over the next few years the one thing that I can promise about technology is going to keep moving and it's going to keep getting efficient we may not have Moore's Law on the processor side anymore but we definitely have it on the data center capability and Cloud

capability side of this we're going to keep marching forward and you're going to need to find a way to train your staff to always be thinking how do I optimize for both agility and cost you want to learn more visit the webpage we've had a ton of conversations that go into these topics in detail from Rob hurstfield at racking the CTO of six feet up just shared about his bill bill versus SAS decision for building a events platform the list goes on and on example as

an example how hybrid cloud is the future for tuning in to my session love to continue the conversation in the chat it will be open for a good half an hour or more have at it you think I'm wrong I can probably see the comments right now you think I'm right I love to have my bias reinforced tell me that I'm right hope you enjoyed the last session in advance with Katina talking about project management it is a must attend it's well edited probably the best

edit video of all the videos but more importantly it is a skill that all it Enterprise Architects and managers should have