Choosing the Right GCE Instance Type for Your Workload (Cloud Next '19)
Hello. Welcome, to our session this. Is choosing the right Google. Compute engine instance. Type for your workload my. Name is Hanna Youssef I'm a product manager on Google compute map on, Google compute engine and and. Today. During this session I'll be walking you through the different virtual machine types available on our platform, as well, as the new infrastructure, announcements, that we're making today during the next 2019. I'll. Focus on describing the different options available to you the capabilities. Of the new instance, types in the new VMs and also. We'll walk through some guidelines, to help you make decisions about the right VM infrastructure. For your workload I hope. That by the end of the session you will have a clearer picture of the different VM, types available and I, hope you will find it easier to pick the right infrastructure, for any work load that you bring to Google Cloud. Before. We start we want to let you know that we do have a Dory for the session so. Please feel free to go to their next application, and add any questions, you have for us we'll, have some time to answer any questions you have in the Dory at the end of the session so. Let's, start by talking about the applications, that you are running today on the cloud I'm. Sure if I ask this room will come up with a list of 100 or possibly, more different, application, types different. Workloads, and different architectures, that you're looking to run on GCP, or on your cloud provider. Some. Of you might be focused, on running high performance, web applications, others. Might be looking for ways to run your ERP systems efficiently, in the cloud or maybe. You're simply, looking for the ability to burst to the cloud and finish a job, that you have at the moment that you're thinking about a bad job. So. How easy did you find the process of matching your needs to the capabilities, provided by your cloud provider. Do. You find it easy to find the, right VM type at the right cost with, the right performance in, the, right locations, that you're looking at do. You find that sometimes confusing. To look through a very, big and large catalog, of instance types with different features and capability. And find, it difficult to compare them I. Bet. You do I personally, sometimes, find it very confusing to look through all of these options available I. Think. One thing that also makes this a more complex question for us to answer is the fact that your workloads are typically not stable what, you start with might, need to change your, performance requirements, might change your budget might change and we. Need to figure out how easy it is for you to make these changes as you get them. We. Believe that this should be simple this. Is why in GCP from day one we really focus on taking a very different approach to. Our instance, types we. Understand, that you need simplicity. So. That your decision making is easy and effective, without, requiring, you to have. A PhD in our catalog or study that for hours, we. Know that you need flexibility, because. Your needs change because.
Your Requirements, come up every day new requirements, come up every day as your, profile workload, or as your business grows and you need different, requirements. Or different resources for example. We've. Also pirate, eyes efficiency. As one thing we always look at when trying to introduce new capabilities, into other infrastructure, and, efficiency. Is very important, over here because on the cloud you shouldn't need to over buy you shouldn't need to over provision you should find something that fits exactly what. You're looking for and what you're gonna benefit from. So. Let's look at this process in more details how, do you actually pick, the right VM shape on Google cloud how. Do you match your exact needs to that available, from our platform we believe. That for 90 percent of, your workloads, this, should be a very simple process where you need to do no comparisons, and you need to find what you're looking for with, exactly one click and this. Is why we've. Really focused on having, that as one of our core pillars as we created our general purpose VM we. Wanted to make them simply, truly. General-purpose, they. Are designed to be simple, flexible. And fit, the majority, of your workload and we. Do that in many ways. First. Of all we focus with our general-purpose VMs, on giving, you flexible, sizing, you. Can simply choose the number, of cores and the amount of memory that you need at any point of time and with, the ratio that you're looking for, you. Don't have to choose from, only, predefined, shapes which. Might. Map to some workloads we have in mind that might not necessarily be exactly, what you're looking for. With. The general-purpose, machine types you can choose more memory more, cores or you can choose a balance and, what's. Very important. In this process is that you don't need to think or worry about pricing. Changing, in strange ways as you change, those ratios, so. By having a very simple, and linear pricing. For, both cores, and memory, that. Becomes one less decision, you need to think about and one less factor you need to consider, it's. One simple pricing that you can have across the board. What's. Also very important, about the January purpose, machine types is that with custom. Machine types with the custom machine types available, on Google cloud platform right.
Sizing, Becomes something that you can actually do on the cloud so, that you don't over provision and you get exactly what you're looking for let's. Talk about custom, machine types. As you. Can see here these. Sliders are actually part of what. I like most about Google, compute engine, simple. Sliders that allow you to just configure, the Machine type to what you have in mind and what exactly if that's what you're looking at. With. Custom machine types you can build an exact fit, and. You. Can prevent any wasted, resources, and. We've. Seen our customers, use this feature to actually benefit from more than 19, percent savings, versus, just using predefined, instances, that are available including. Predefined. Instances, available, on Google compute engine itself. Another. Great way that custom machine types help you is that, it provides you with something, that we call extended, memory and extended. Memory is really meant to break any barriers, you have to creating the exact shape you're looking at so. For example let's take an example of a workload that has very, expensive per core requirements, on, any, clouds you typically have to, choose. The memory that you're looking for and you. Get a predefined, amount of course available with that and then you will have to pay the licensing, for all your course you don't have any choice, with. Extended, memory you can actually figure out the amount of memory that you need and then, decrease, the number of course that you're buying with, the VM so that you don't have to bake to pay for extra licensing. That you don't actually need another. Great. Thing about custom, machine types and about the general-purpose VMs, is that. We're also. Aside. From giving you the ability to create the exact fit shape. That you're looking for we, also try to help you with that decision and help you with figuring out what the right number and the right fit is. We. Do this through the right sizing, recommendations. So. If you've been on the Google cloud console. Recently, you might have seen a screen that looks like this one and what. The screen shows you is our. Own recommendation. And the, platform intelligence, that we make available to you so, that you can figure out if what you're running today is really, the most efficient, way to run your workload, on. The. Top here for exam I have an 8 v CPU VM, that I'm not fully utilizing and because, of that Google, compute engine right away gives, me a recommendation that, I can actually downsize. To a different VM and directly. Save more than 150, dollars per month. In. Many ways this, is meant to give you the ability so that with custom machine types and with the right sizing recommendation, you, get the exact fit and you, avoid overspending, on infrastructure. Resources that you don't need. Another. Thing which is really significant, and great. About general-purpose, VMs, is that. They give you a lot of flexibility, to change, the. Attachments, that you add to those VMs a great. Example, is storage so. On, a typical day you might think. Coming, into the cloud that, you need locally, attached storage to be able to get the performance that you're looking for maybe. A few days later after some performance, testing you see that you actually don't need all, that performance and you, prefer to have something that's network attached and. More. Durable for your, critical, data. In. A typical case you would have had, to change your entire VM, type choose a different one that has local, storage or not local storage with. Google compute engine we try to make that a much, easier process, so. That you can just have a VM running, on GCE and go, and change the, storage mode or the storage type that you're using with this VM as you, can see here you can choose the local SSD you can choose persistent. Disk which is network attached and you can even choose among the different types of persistent. Attach. These are available to you. The. Main goal of thinking, about our infrastructure. This way and presenting, it this way is that. We simplify, the process of decision making and we simplify, the process of operations, for him. This. Applies to a lot of things across Google compute engine a couple, of examples. Have. To do with how we manage CPU, platforms, and GPUs. We simply believe that there are decisions you shouldn't have to spend time on and there, are things that are intelligent cloud platform, should be able to do automatically, for you, one.
Clear Example is, managing, the hardware upgrade. Cycle. Maybe. For some of your workloads, you do care about understanding. The underlying CPU architecture, and optimizing. Your code for it you, might be using certain instruction, sets for example. However. From our experience, we've, seen that most of the time you actually don't need to worry about that and you don't need to think about it, you just need something the trend and runs well on the cloud and for. These cases we, give you the ability to choose, a minimum CPU platform, as you. See in the screen under, CPU platform, this is intel skylake or, later and what. That means is that by, simply choosing, this box and choosing. That option you. Give us the ability to manage the hub the, hardware upgrade cycle on your behalf so that we can continue to move you to the more recent infrastructure. More, efficient, infrastructure, as this becomes the, most appropriate one for for. Us to use and for you to use without, you having to go and do any work you, don't stop the instance you don't recreate, another one, it simply, upgrades on your behalf. If. We're talking about flexibility. Another, thing that we, believe helps you with general-purpose, VMs, is the, ability to choose different accelerator, types. You. Might have been using K ATS for example at a certain point of time but, now want to experiment with the new three for GPUs, you. Don't need to go and rebuild your entire infrastructure, again. You can remove on GPU and add another and you're set to go, without. Having to go back and figure, out if for example the, perfect shape that you've been recommended also exists, with the GPU instances. So, to summarize, for the january purpose VMS we, covered how a wide variety, of. Options. And of workloads, are available, with those general purpose VMs we, talked about how they enable you to run things that, might be very different and have different resource, requirements, by, changing, the amount of cores, and memory. That you have available. Across. Different workloads from databases to applications, mobile gaming and dev fantasy. Environments, we. Also discussed, the great flexibility, you get with general-purpose, VMs, you, don't have to spend a lot of time trying to research or compare, the instances, you. Get custom machine types which enable, you to create an exact fit for your workloads and even. If finding the exact fit is not a right task you get recommendations, that enable you to do that this. Makes general-purpose, VMs, great for a wide variety of workloads, and makes, them really a very, simple way to get started without having, to rethink or to look through a very large catalog, of instance, types options, and compare them. We've. Seen our customers make great benefit, of these features and these capabilities and tell. Us a lot of times how custom machine types really enabled, them to change their economy, to change the economy of running their infrastructure, and spend. Their, money and their focus on their. Operations, on or 3d matters. Now. Let's talk about bursty, workloads. Many. Google cloud customers look, for a better option to run more cost sensitive and bursty, jobs when. They need to. This. Can be the kind of job where you just need to. Run. Thousands, of cores for example for a very short period of time on the cloud so that you complete a certain task you have in mind a good. Example here is rendering. Jobs for example we. See a lot of Studios coming, to GCP so that they run a long running a big rendering job get the tasks done in a very short period of time by leveraging thousands. Of yen's at one time. Typically. These workloads only, care about the time it takes for them to complete they. Might not essentially, care about this being a real-time process where they are interacting, with their customer in every step of the way and. I'm. Sure that you would agree that economics, become a very big part of this decision when, you're looking to run tend of tens, of thousands, of VMs for a very short period of time you really don't want to be spending too much for that. For. These cases, preemptable. VMs, is GCPs. Recommended, solution.
Preemptable. VMs are made for batch check. Pointed, and high throughput computing. They. Are simply, short running VMs that. Can run for up to 24, hours and can, be preempted, at any point of time and. With. Those constraints, they, become 80%. Cheaper for you to run. The. Great thing about these vm types is that they are actually not a separate VM type it's just a flag that you can add to any instance, at any point of time you don't, have to go and redo your research or think about some. Capabilities, you might be missing if you chose preemptable VMs for example. This. Makes them ideal for a variety of workloads, from, genomics, to. Computational, chemistry to, data processing, and, simulations. The. Greatest thing about them is that you get directly. 80%. Of which. Is a fixed discount. You. Predictable, pricing, without. Having to face. Any unpredictability, into this process as long as you have jobs that are full tolerant, and can manage preemption. You, directly get this benefit. Again. We've seen many GCP customers, make great, use of this capability, many. Of them run tens, of thousands, of course and tens of thousands, of VMs while, saving money an equation. Which is typically not very easy to solve. Now. Let's move to a different workload, let's. Talk about gaming, as an example. Multiplayer. Games. Specifically. Are a great example of a, computer intensive workload. We. See typically triple-a titles looking for the fastest, and best experience. They can present to their end customers. For. These companies frame, processing, time is critical if, that, ended up happening slower, than expected than, their, players the, gamers would actually notice that, their game is not performing, as expected and, it would impact their, experience. This. Is why we've been working for a very long time on. Introducing. The new compute, optimized VMs which, is one of the new instance, types that we've announced today during next 2019. Compute. Optimized VMs, simply provide, the highest per core performance, on GCP. They. Are the best fit for compute, intensive workloads, that might include triple-a, gaming, high-performance. Computing. And a. Lot of very high performance. Websites and applications. These. Instances, provide, more, than 40% higher, performance. For. Workloads, that are compute, bound and care about single threaded perf. What's. Great about those VMs is that with, this new introduction, they are actually, powered, by the latest CPU. Architecture, available on the market so, these are powered by the new Intel, scalable, processors, the second generation codenamed. Cascade Lake and they, offer the highest frequency, available. On DCP with. Up to three point eight gigahertz, in all, quarter, ball. Let's. Talk a little bit more about compute, optimized VMs. As. You. Can see they, already come with multiple shapes and oceans so. Today they arranged from 4 to 6 TV CPUs, from. 16 to 240. Gigabytes, of memory offering. You a 4, to 1 gigabyte. Per, vcp ratio. They. Also enable you to manage. Or use multiple. Storage options, so, you can use network attached storage persistent. Disk or you, can use local SSD local. Storage as with any other GC, instance. We're. Also happy that we are introducing, these VMs, with. The new networking, caps that we have available that were announced today so, they offer directly, up to 32, gigabits, per second. And. Going, beyond the shapes and the hardware we've. Really focused on making the computer optimized, VMs, the. Best option. For you when you have new real-time and. Highly. Demanding, workloads. So. These VMs you get full transparency into. The underlying hardware. You. Get explicit. Numa visibility, so that you get the best performance across.
The. Different VM pipes and different VM sizes, you. Also get very tight, consistency. And. Very. Predictable, performance so, that you can bet your most intensive, applications, on those VMs. For. Those of you that are looking to do more, fine-tuning, of your performance, you, also get access to things such as, the. Processor, power state so, that you can find, you in your application, performance, on those feeders. So, to summarize, what we covered for compute, optimize VMs these are one of our new introductions. To DCP as we, look to cover more, of the edge cases that you might have for your different workloads, they. Have the highest performance, per core available, on Google compute engine, they. Are powered, by the latest intel's. Second-generation, scalable, processors, and go up to three point eight gigahertz. In all quarter, both in. Addition they have minimal. Variability. Very. High consistency and, offer. You multiple shapes, so, that you can find the one thing which is looking for and fits your workload, this. Makes them a great fit for multiple, applications including. Multiplayer, gaming, high-performance. Computing, video, encoding, real-time, web serving and others. Now, let's move to some of the other highly, demanding workloads different. Types of highly demanding workloads. One. Great example here, might be. Databases. Specifically. Hi and databases. We. See a lot of customers running large. ERP systems for example a zippy ERP, systems, and. What. They are looking for it typically is the, fact that as their database continues, to grow they, are looking for an infrastructure, that will grow with them and will, be able to scale with their needs, while. They think about this they are also looking for an infrastructure, that continues, to give them the benefits of running on the cloud the flexibility, that comes with that and the, ability to have higher fault, tolerance and many, of the great features that we talked about. This. Is where the memory optimized, VMs come into the picture and, the. Way we really define memory, optimized, VMs on Google compute engine is that, they are the large memory VMs, one, terabyte and higher. These. That typically, tend to be multiple, socket systems very large systems, and tend. To have a very, high amount of memory per core so, that they can continue to scale with your memory needs. Memory. Optimized VMs, are built for. Memory intensive workloads. Including. In-memory databases, and in-memory, analytics. They. Provide the highest gigabyte, per core available, on GCP with options, that range from 15. To 24, or 25 gigabytes, per V CPU and they. Also offer you the lowest. Dollar. Per gigabyte available, with these configurations. Multiple. Shapes are available for you under, the memory optimized VM. Category. Ranging, from one. Point five terabytes up to. The sixth and twelve terabyte VMs that we just announced today and with. Plans to even extend up to 18 terabytes and more. If. You're thinking about how the infrastructure, continues to scale with your database and scale, with your needs then, one of the things we've really, had. As one of our guiding principles, for those PMS is the, speed at which we've been trying to add new sizes and continue to scale with your requirements, so. You can see here that in approximately. The last 18 months those. VMs have increased, in sizing by more than 10x for the number of course they provide and 60x, for the amount of memory available on, those VMs.
My. Favorite part about memory, optimized VMs though is that they, continue to give you all the benefits of Google compute engine that, we've just described. So. They are fully flexible, and they, provide you with very high availability. Let's. Talk about availability. With. Memory optimized VMs you continue, to get live migration. Supporting. Zero downtime for your most critical applications. And for. Those of you not familiar with live migration it's, simply, our way of making sure that maintenance, events are transparent. To your workloads. When. Google Cloud expects. That, or, sees them that the maintenance event, is about to happen, we, transparently. Spin. Up another version of your VM copy, all the data over so. That you don't have to experience any downtime or impact, on your application, and this. Continues, to be one of the things that we keep as a core part of infrastructure, as we expand to more sizes. Another. Thing here is flexibility, if, you're, running the type of deployment and the type of workload where you expect your requirements, to change months, over months then. You definitely don't, want to be stuck. Into a reserved, size, we're. Moving to a different size requires, a different contract, or a different commitment, and, this, is why the commitments, that we provide for memory optimized VMs are really, just for memory optimized VMs you, can use the same commitment, for one for. A one terabyte VM you can, add to that and use it for four terabyte, VMs or you can divide it across different VMs, as you scale out or scale up it's really up to you it's, one big category, that you can divide it the way you see fit as you scale up or as you scale down. Another. Great thing about memory optimized, VMs is the, simplicity, these. Are native gcpd. Arms native, Google compute engine VMs, they. Fully support all, the, Google compute engine features, and capabilities, you. Use them directly from the console from. G-cloud, from, the API you can use labels you can use any other features, that you're using today including. The policies, for example that you have defined for your organization's. And. Manage, them exactly as any other small VM available, as part of your deployment. The. Last part and mention, about the memory optimized VMs is our focus on making sure that we introduce new innovation, as we continue to expand our sizing, a great. Example of that is, the support for the latest memory technologies, and, you. Might have recently seen some announcements, from Google compute engine where where, we talked about Intel, update DC persistent, memory. This. Technology, is specifically, useful for those of you who have been seeing a gap between. The, cost, and the performance, available, from DRAM and that, available, from flash, or from SSD. If. You have applications, that, require better, performance. Than storage, but. Cannot. You cannot afford to have them running on DRAM due to the cost associated with that then, this fits in into that gap over there as you can see it, fits in that place between, memory and storage. The. Other great thing about this technology is that it also gives you the availability or the ability, to have persistent. Memory so. That for certain applications, the, application, going down doesn't, mean that you actually lose all of your data even, though you had get the performance or a performance, that's very close to that that you get with DRAM. Google. Cloud has done a lot of early work with this technology and focused. On availing it first so that you can test it on our platform. With. That we give you much higher density per. VM than. That which can only be available with, the regular memory technologies. Approximately. 4x, more memory, at. A significantly. Lower cost. It. Tends to be much faster and lower latency, the, storage and is, a great fit for in-memory, databases, across the board.
Today. We do have in our alpha, testing programs, a few, shapes available where you can test, Intel. DC persistent, memory for yourself, and add. With. A small note with, the 96, V CPU note up, to, 6.7. Terabyte. Across. The, RAM and obtain, DC, persistent, memory. If. You have workloads, where you don't need the performance, of duran and. You, want to leverage the TCO improvements, this might be a great fit for your testing. So, let's summarize the different instance types we've covered. We. Thought how we. Talked about how with Google compute engine, you get very flexible, general-purpose, VMs. The. General-purpose VMs, tend to be the. Most flexible and the most simple to use they. Are great for a wide range of workloads, across web serving databases. Gaming, and others. They. Also give you the ability to right-size, the VM to exactly, what you're looking for, you get recommendations that, help you with making sure that you're picking the right size they. Are very flexible and enable. You to directly, attach or did the app or detach. Accelerators. Or storage options. Then. We covered the compute optimized VMs, one, of the new things that we're introducing on Google compute engine for your. Very performance. Intensive, workloads. Compute. Optimized VMs, offered, the highest performance. Procore available. On GCP, they. Are a great fit for triple-a, gaming HPC. High-performance. Websites and some types of scientific modeling, they. Are available in multiple shapes, and, enable. You to get approximately, 3.8. Gigahertz, in all quarter, Bo. We. Talked about memory, optimized, VMs and these. VMs are a great fit for your memory intensive, workloads. They. Have a lot of sizes that go from 1 terabyte to approximately, 12 terabytes today, with, more coming, they. Are a great fit for high end databases, in memory, databases. As well as real-time and in-memory, analytics. They. Also offer you great flexibility. Since. You can always scale up or scale down, or, divide. Your. Commitments, across them in different shapes and sizes. We. Also talked about preemptable, yams which. Is really a simple flag that you can add to any of the Empire across, the different types we talked about regardless.
Of Your CPU, and memory, and performance requirements and once. You add that flag you automatically, get the most, economic, option for, bursty fault-tolerant. Workloads. This. Is a great fit if you're running a rendering, job, a bad job or some simulations, that you just need to compute for but. You don't care if it gets preempted, and then continues, to run again and you, have that kind of intelligence, into your platform. So. To, summarize, today. We really focus on how we can make some of your decision-making simpler. On Google compute engine by. Making sure that we always continue to focus on simplicity, flexibility. And efficiency. We. Talked about how you can use simple guidelines, to. Make, simple choices on whether you just want a wide range of flexible, options with, general-purpose VMs, or you're. Looking for the most performant. Option available with, compute optimized, VMs. We. Also covered, have how for some of your very memory. Intensive applications. With. Memory optimized shapes you, get the benefits of the cloud with. The sizing, that, you're looking for and you. Get a lot of benefits related to availability and flexibility. Even as, you go to these large sizes we, hope that give that that gives you a general overview into the options available for, you and makes it a little bit easier to decide on what you're looking for on Google compute engine.