Amazon ElastiCache Serverless | Amazon Web Services

Key Takeaways

Hello and welcome to this short snackable video on Amazon Elasticash Serless. My name is Paul. I'm a specialist solutions architect for in-memory databases on AWS. Let's get started. Okay, we'll start by exploring some typical challenges you might encounter with caching solutions, both self-managed options and when using nodebased Elasticash clusters. Then we'll introduce Amazon Elasticash serverless. We'll go over features and benefits. We'll learn how the service provides high availability and how it scales to meet workload demands. Then we'll see how this all comes together in a short demonstration. Finally, we'll explore the pricing model. When teams build their own caching solutions from scratch, they take on significant operational challenges. This means managing all the infrastructure themselves, building monitoring systems, and handling complex tasks like scaling and failover. They're also responsible for capacity planning, security updates, and regular maintenance, all of which require specialized expertise, and considerable time investment. Elastic nodebased clusters provide teams with a proven, reliable caching solution that eliminates many operational headaches. While you'll still make key decisions about node types and scaling to match your workload needs, AWS handles the complex infrastructure management for you. Like any provision service, you'll need to carefully balance capacity. Overprovisioning means paying for unused resources while underprovisioning could impact application performance. You maintain control over important aspects like maintenance, scheduling, and configuration settings, allowing you to optimize your cache environment for both cost and performance. This balance of control and automation makes node-based elastic clusters an excellent choice for many teams. Though for those seeking to eliminate capacity planning altogether, there is an even more automated option available. Amazon Elasticache serverless allows you to create a cache in under a minute and instantly scale capacity based on application traffic patterns. The service is compatible with memcachd reddis oss and valky and is highly available by default. Let's look at the features and benefits in more detail. Creating an elastic cluster on AWS has never been easier. It literally takes a minute. There's no capacity management to worry about. And as you'd expect, the service is fast. It provides microscond response times for your applications. Let's look at how Amazon Elastic Serverless simplifies the developer experience. Everything starts with a single end point. Developers can focus purely on writing code while the service handles all the underlying complexities. Operations teams can rest easy knowing that multi-AZ replication automatically ensures high availability. Software updates and minor version upgrades are managed automatically, eliminating another operational burden. The scaling is equally seamless. The service uses bursting techniques to rapidly respond to additional workload demands. And as your data grows or your workload changes, the service automatically adjusts capacity by adding or removing shards, all without any manual intervention. Let's consider the architecture in more detail. The service delivers a seamless caching experience capable of handling millions of requests per second with submillisecond latency. Let me explain how it works. At its core is an innovative proxy layer built using Rust for optimal performance and security. Your application connects through a single endpoint to the service VPC where the proxy intelligently manages all traffic. The architecture has been designed to span multiple availability zones using smart routting through route 53 and a network load balancer to maintain ultra low latency. How has Elasticash serverless been optimized to scale quickly and efficiently? Firstly, highfrequency polling. These systems check each shard every few seconds, ensuring real-time responsiveness rather than relying on delayed metrics. The service also makes use of warm compute pools, preconfigured nodes with all the necessary software installed, ready to join the cluster instantly when needed. This dramatically reduces scaling response time. Heat management is an intelligent approach that triggers scaling actions during peak pressure periods. And finally, optimized slot migration enables the transfer of millions of keys quickly and atomically all while maintaining normal operations. The service employs a dynamic resource management approach, moving away from fixed memory and CPU allocations. The key to this efficiency is intelligent overs subscription, allocating resources based on actual usage patterns rather than peak capacity requirements since most tenants rarely need their maximum allocation simultaneously. This innovative approach enables instant scaling in both directions. When workloads demand more resources, the system immediately provides extra memory, CPU, and network bandwidth. It starts with vertical scaling for an immediate response, followed by horizontal scaling as needed. The platform continuously monitors resource utilization, ensuring optimal performance while maintaining cost efficiency. For horizontal scaling, there are three key stages: detection, provisioning, and data rebalancing. Detection occurs within seconds to meet immediate workload demands. The service also employs predictive scaling, analyzing usage patterns to anticipate resource needs in the coming minutes. This proactive approach ensures smooth handling of sudden traffic spikes. Provisioning is remarkably fast, taking just a minute through the use of warm pools. These are preconfigured cache servers ready to deploy instantly. This allows the service to rapidly expand cluster capacity when needed. Let's look at how Elasticash serverless handles data rebalancing during scaling operations. The service performs parallel slot rebalancing, enabling rapid cluster expansion. It continuously monitors usage at the slot level, identifying hot slots that require attention and intelligently redistributes data to maintain balanced load across all shards. The system uses advanced batching mechanisms and optimize network buffers to efficiently transfer data between shards. This sophisticated approach enables the service to double cluster capacity in minutes while maintaining consistent performance all without any operational overhead for your team. Balky is an open- source high performance key value data store. The project is backed by the Linux Foundation ensuring it will remain open source forever. Elasticash serverless with Valky 8 brings remarkable scaling improvements. The service now scales from zero to 5 million requests per second in under 13 minutes, doubling capacity every 2 to 3 minutes. We've covered a lot of theory about Elasticash serverless. So now let's put it into action with a real world demonstration. I've built a simple weather application with a NodeJS backend that uses bulky glide. The application sits on compute instances behind an application load balancer and all the weather data is stored in elasticashe serverless for valky. First I'll show you the application and a typical API response. Then we'll conduct some load testing to see elasticasheach serverless in action as we scale up the traffic. Let's look at our weather application which displays a 7-day forecast interface. This demo showcases how Elasticash serverless with Valky delivers real-time data through API integration. The application responds instantly to user interactions, whether browsing different cities or refreshing forecasts. Behind the scenes, each forecast day corresponds to a unique key stored in Valky, ensuring efficient data organization and retrieval. We've included weather data for 1,00 different cities. The forecast is machine generated for demonstration purposes. Our focus here is on testing performance capabilities, leveraging features like pipelining and read from replica to maximize throughput. Looking at the API response, we can see weather data for London covering a 7-day forecast. Each element in the forecast array comes from a pipelineed request. The service executes eight get operations, one for each day, plus additional metadata using pipeline mode with read from replica enabled. And the performance is excellent. The entire operation between the web server and the cache completes in just 1 millisecond, demonstrating the exceptional speed of Elasticash serverless when using Valky. Let's look at our load testing results. Running two JMeter tests simultaneously, we achieved excellent performance metrics. The first test delivered 38,000 API requests per second with an average response time of just 7 milliseconds. To simulate a burst in traffic, the second test added another 30,000 requests per second, bringing our total throughput to just under 70,000 API requests per second. Since each request performs multiple cache operations at peak, this translates into nearly half a million cache operations per second with approximately 30 million operations per minute visible in Cloudatch. Let's examine these performance graphs. You can see the clear spike in traffic when we launch the second test. What's particularly impressive is the consistency in latency under 500 microsconds even as the load increases. And here's the key takeaway. With Elastic Serverless able to scale from zero to 5 million requests per second in minutes, you can be confident that caching won't be your performance bottleneck. Now, let's move on to the pricing model. With Elasticash serverless, you pay for data stored in gigabyte hours and the compute used by your application workload in Elasticash processing units, otherwise known as eCPUs. Amazon Elastic for Valkyrie starts as low as $6 a month on serverless and offers a 33% saving compared to other supported engines. ECPUs are a unit that includes both vCPU time and data transferred. Reads and writes require one eCPU for each kilobyte of data transferred. For example, a get command that transfers 3.2 kilob of data will consume 3.2 two eCPUs. Commands that require additional virtual CPU time or transfer more than 1 kilobyte of data will consume proportionately more eCPUs. If you'd like to learn more about Elasticash, follow the links on screen. Thank you for joining me in exploring how Elasticash serverless simplifies caching while delivering enterprisegrade performance and reliability.