At Re:Invent 2023, Peter Desantis, senior vice president of Amazon Web Services, announced three new serverless products in his keynote address, spanning its database and analytics portfolio, with epic upgrades that enable customers to scale their data infrastructure faster and easier to support demanding business use cases. This move reaffirms AWS's continued commitment to "innovating for customers" and reinforces its leadership position in the serverless space.
The status of Amazon Web Services cloud computing giant was not achieved overnight, which is inseparable from its long-term crazy "self-rolling" and unique reverse work principle. Different from the competition model of technology PK and performance rankings, Amazon Web Services always focuses on the needs of customers and continues to innovate for customers, which is also the key reason why enterprises are willing to continue to pay for their cloud services. This time, what innovative features does Amazon Web Services' three new serverless products bring to customers?This article is the first thing for you!
【Release】The new Amazon Aurora Limitless Database
When it comes to database management, enterprises are often challenged with physical server resource constraints. Managing a sharded database involves a lot of complex operations, and the need to manage dozens or even hundreds of shards to achieve massive scalability is an increasingly difficult task when it comes to making changes to events across shards.
Amazon Aurora Limitless Database, the first major serverless product of the conference, implements the automatic scaling of Amazon Aurora relational databases, completely solves the workload of enterprises to manage shards, and helps enterprises easily extend their databases beyond the write throughput of a single server. For enterprise database administrators, they no longer need to manually plan database splitting, nor do they need to worry about the impact of splitting operations on database performance, everything is automated.
Automatically scale high-throughput workflows to millions of writes per second based on your customer's data model and manage petabytes of data in a single database
Caspian provides resources for shards, allowing them to scale up and down as needed
Grover makes it easy to clone and repartition databases, allowing enterprises to efficiently route queries to the appropriate shards, enabling fast distributed data exchange across all shards.
Obviously, the amazing ability to scale infinitely and the performance of the database is inseparable from the support of two underlying technologies - Caspian and Grover are no strangers to users, they have existed in the use case scenarios of the past, and this time they have ushered in an innovative change.
grover
As an internal application of Amazon Aurora, Grover separates the database from the storage itself, making the log itself a database, which can significantly reduce IO, and has innovative capabilities such as remote processing, automatic replication, and seamless scaling.
Instead of logging locally, grover replicates each log to multiple Availability Zones to ensure log availability and durability
Grover not only stores logs, but also processes them, creating identical copies of the database's internal memory structures on the remote system and sending those data structures back to the Amazon Aurora database whenever needed, significantly reducing the number of iO on the primary database
Gover provides durability across multiple Availability Zones without the need to replicate databases. If the Amazon Aurora database fails, or an entire Availability Zone fails, the customer can restart the Amazon Aurora database in a different Availability Zone
Enterprises can also enable serverless scaling of database storage. Because every Amazon Aurora database has access to Gover's multi-tenant distributed storage service, enterprises can seamlessly and efficiently scale from a single table to massive databasesWhen the database becomes smaller, enterprises can abandon some nodes and stop paying to reduce costs.
Compared with traditional databases, Amazon Aurora no longer needs to write persistent memory, and users only need to log in to Grover to write logs efficiently with a small amount of IO. It is worth mentioning that Grover has reduced the I.O requirements of Amazon Aurora database storage systems by 80%, and customers can achieve 3-5 times the cost performance compared to open source managed databases with Amazon Aurora.
caspian
As a virtual hypervisor, Caspian includes a new hypervisor, thermal management planning system, and improvements to the database engine itself, allowing Amazon Aurora Serverless databases to resize in milliseconds in response to changes in load and efficiently allocate resources, revolutionizing database management.
A caspian instance is always set to support the maximum memory available on the host on which it is running, and these resources are not allocated to the hypervisor on the physical host, instead, the physical memory is allocated separately based on the actual needs of the database that Caspian is running, a process controlled by the Caspian thermal management system
Caspian can run multiple databases and allow them to efficiently share underlying host resources, with the security and isolation of a hypervisor;
When the database needs more memory, the Caspian thermal management system is responsible for managing the resources of the underlying physical hosts, scaling flexibly and without impacting performance when the database is migrated.
As Peter demonstrated at the conference, Caspian is constantly trying to optimize the database set by constantly scaling up and down as the load changes. This constant dynamic ensures that the heat is balanced across the fleet, ensuring the efficient and stable operation of the underlying infrastructure.
【Release】The new Amazon ElastiAche Serverless
Obviously, Amazon Web Services has provided great help to enterprises in terms of serverless capabilities of databases, but the challenges faced by enterprises are far more than these, and data caching is the first oneTwo difficult problems that need to be solved urgently. The performance of caching is very dependent on the memory of the hosting server, which is not serverless;In addition, largeThe amount of resources being used to cache data will incur high costs, while insufficient resources can lead to the loss of valuable data.
That's what Amazon ElastiCache Serverless wants to solve. As the second new product announced by the conference, it provides a highly available and scalable serverless caching service. As a serverless option for Amazon ElastiCache, enterprises can launch Redis or Memcached caching solutions based on this without the need to provision, manage, scale, or monitor node fleets. In just one minute, customers can create caches and instantly scale capacity based on application traffic patterns.
Under the hood of Amazon ElastiAche Serverless is a sharded caching solution that is very similar to the technology that powers Amazon Aurora Limitless Database;
Caspian, the underlying compute layer, can resize shards and scale up and down, which helps resize caches to take full advantage of performance and cost;
Improvements to the request routing layer enable multiple cache shards to receive data from a single cache endpoint with extremely low latency;
with redis 7 and memcached 16 Compatible, with an average lookup latency of half a millisecond, and supports up to 5 TB of memory capacity.
Distributed databases provide great convenience for enterprises to manage data efficiently, but time synchronization problems arise one after another. In a distributed environment, the clocks of different nodes may not be fully synchronized, and clock skew may lead to inconsistent order of log entries, which will seriously affect transaction processing and data consistency.
amazon time sync service
Five years ago, Amazon Web Services launched a clock synchronization service that enables Amazon EC2 instances to be accurate to milliseconds. This time, Amazon Web Services has achieved a further innovation - Amazon Time Sync Service, a high-precision time synchronization service, has been launched. This is a server clock that can be synchronized and can create an ordered event log for the database.
Powered by redundant satellite connections and atomic reference clocks in Amazon Web Services Regions, current time readings of the UTC global standard are available with microsecond latency
Custom time-synchronized private network, device integrated Nitro chip and FPGA, combined with the Nitro chip in the Amazon EC2 host, enables timing pulses to be distributed directly to each Amazon EC2 server. These distribution steps are all done in hardware, avoiding variables caused by drivers, operating systems, or network buffers
Enterprises can take advantage of Amazon Time Sync Service with supported Amazon EC2 instances, making it easier to sequence application events, measure single-item network latency, and dramatically increase the velocity of distributed application events
Based on the Amazon Aurora Limitless Database, the system can support hundreds of thousands of order events per second.
【Release】The new Amazon Redshift Serverless
Finally, we see data warehouses designed for massive amounts of data. As your data warehouse processes millions of queries every day and serves a large number of users, a mechanism to easily configure and scale data warehouse capacity based on query volume works well if all queries are similar;But in the absence of unified queries, sometimes large and complex queries can slow down the system and affect other smaller queries. As a result, businesses want to manage their data warehouses faster, better, and more cost-effectively.
In the past year, artificial intelligence has ushered in disruptive innovation, and Amazon Web Services has never missed the trend of technological development on the road of customer innovation. The finale of the conference, Amazon Redshift Serverless, features AI-driven scalability and capability optimization to find the best way to run queries by analyzing each query and considering query structure, data size, and other metrics to find the best way to run queries, taking into account efficiency, impact on the cluster, and more.
A machine-based Xi model that enables future workload patterns and adjusts resource capacity in advance
A real-time query analyzer that uses machine Xi to estimate the resource requirements of each query and allocate them appropriately. The system analyzes more than 50 unique features for each query
Optimize each query to reduce costs or improve performance based on customer needs. Queries have different scaling modes such as linear, sublinear, and superlinear;
Optimize for different types of large, small, and complex production workloads to deliver up to 10x better price/performance.
It is worth noting that in DrIn Werner's latest presentation, Serverless Computing's lightweight virtualization Firecracker was again mentioned, which uses KVM's new virtualization technology to allow customers to spin up lightweight microVMs in non-virtualized environments in less than a second, taking advantage of the security and workload isolation provided by traditional virtual machines, as well as the resource efficiency of containers. All of these features and services are a testament to the tremendous innovation of Amazon Web Services in the serverless space.
In the continuous commitment of "always innovating for customers", Amazon Web Services has never stopped exploring. What other innovative surprises are there in the future?We'll see.