AWS Database Blog

New for Amazon ElastiCache for Redis 7: Get up to 72% better throughput with enhanced I/O multiplexing

Amazon ElastiCache for Redis is an in-memory data store, delivering real-time, cost-optimized performance for modern applications. It is a fully managed service that scales to millions of operations per second with microsecond response time.

Open-source Redis (“Redis OSS”) is one of the most loved NoSQL key-value stores, and is known for its great performance. Our customers choose to run caching on ElastiCache for Redis to help reduce database costs and help improve performance. As they scale their applications, they have a growing need for high throughput with a high number of concurrent connections. In March 2019, we introduced enhanced I/O, which delivered up to an 83% increase in throughput and up to a 47% reduction in latency per node. In November 2021, we improved the throughput of TLS-enabled clusters by offloading encryption to non-Redis engine threads. In this post, we are excited to announce enhanced I/O multiplexing, which further improves throughput and latency, without requiring you to make additional changes to your applications.

Enhanced I/O multiplexing in ElastiCache for Redis 7

ElastiCache for Redis 7 and above now includes enhanced I/O multiplexing, which delivers significant improvements to throughput and latency at scale. Enhanced I/O multiplexing is ideal for throughput-bound workloads with multiple client connections, and its benefits scale with the level of workload concurrency. As an example, when using an r6g.xlarge node and running 5,200 concurrent clients, you can achieve up to 72% increased throughput (read and write operations per second) and up to 71% decreased P99 latency, compared with ElastiCache for Redis 6. For these types of workloads, ElastiCache for Redis 6.2 processing can become a limiting factor in the ability to scale. With enhanced I/O multiplexing, each dedicated network I/O thread pipelines commands from multiple clients into the ElastiCache for Redis engine, taking advantage of ElastiCache for Redis ability to efficiently process commands in batches. Enhanced I/O multiplexing is automatically available when using ElastiCache for Redis 7, in all AWS Regions, and at no additional cost. See the ElastiCache user guide for more information. Once you’re using ElastiCache for Redis version 7, no application or service configuration changes are required to use enhanced I/O multiplexing.

Let’s drill down a bit more and examine how it works. The following diagram shows a high-level view of how Redis OSS handles client I/O, up to ElastiCache for Redis 6.

The Redis OSS engine serves multiple clients simultaneously. The engine is single threaded, and the processing power is shared between connected clients. For each client, the engine handles the connection, parsing the request and running the commands. In this case, available CPUs are not utilized.

The next diagram shows ElastiCache for Redis with enhanced I/O (released for Redis OSS 5.0.3), where available CPUs are used for offloading the client connections, on behalf of the ElastiCache for Redis engine. A similar approach was taken in open-source Redis OSS 6, which introduced a similar concept called I/O threads.

With enhanced I/O, the engine processing power is focused on running commands and not handling network connections. As a result, you can get up to an 83% increase in throughput and up to a 47% reduction in latency per node. With multiplexing, instead of opening a channel for each client (as can be seen in the preceding diagram), each enhanced I/O thread combines the commands together in a single channel with ElastiCache for Redis engine, as shown in the following diagram.

In the preceding figure, multiple clients are handled by each enhanced I/O multiplexing thread, which in turn uses a single channel to the ElastiCache for Redis main engine thread. The enhanced I/O thread collects and multiplexes client commands into a single batch that is forwarded to the ElastiCache for Redis engine for processing. This technique allows for running commands more efficiently, resulting in improved speed, as we demonstrate in the next section.

Performance analysis

In this section, we present benchmark results for the memory-optimized R5 (x86) and R6g (ARM-Graviton2) nodes. We chose a typical caching scenario that included 20% SET (write) and 80% GET (read) commands, with a dataset initialized with 3 million 16B keys with 512B string values. Both ElastiCache and the client applications were running in the same Availability Zone.

Throughput gain

In the following chart, we share results from benchmark runs exercising a range of client concurrency (400 clients, 2,000 clients, and 5,200 clients) against different node sizes. The graph captures the percentage improvement in throughput (requests per second) that enhanced I/O multiplexing delivers.

With up to 72% gain in requests per second, enhanced I/O multiplexing introduces throughput improvement across the board, as can be seen in the preceding graph. It’s important to note that the more client load on the system there is, the higher the improvement is. For example, r6g.xlarge shows 26%, 48% and 72% ops/sec gain with 400, 2,000 and 5,200 concurrent clients, respectively. The following table lists the raw numbers demonstrating the improvement.

ElastiCache for Redis Version 400 Clients 2,000 Clients 5,200 Clients
6.2.6 (ops/sec) 313,143 235,562 202,770
7.05 (ops/sec) 395,819 348,550 349,239
Improvement 26% 48% 72%

The multiplexing technique improves ElastiCache for Redis processing efficiency, similar to the Redis OSS pipeline feature. For both Redis OSS pipelines and the ElastiCache for Redis enhanced I/O multiplexing, the commands are processed in batch. However, while Redis OSS pipeline batch includes commands of a single client, ElastiCache for Redis enhanced I/O multiplexing includes commands from multiple clients in a single batch.

Client-side latency reduction

Maxing out throughput is not recommended and results in CPU utilization spikes and response time slowdown. To avoid that, we tested latency at 142,000 requests per second, which was 70% throughput utilization of ElastiCache for Redis 6.2.6 on r6g.xlarge. We measured client-side latency with ElastiCache for Redis 6 and compared it with the same load on ElastiCache for Redis 7. The following table shows the results of a test running on r6g.xlarge. It captures the percentage of P50 and P99 latency reduction.

.ElastiCache for Redis Version P50 Latency P99 Latency
6.2.6 (ms) 0.55 8.3
7.05 (ms) 0.44 2.45
Latency Reduction 20% 71%

Similar to the throughput gain, the client-side latency reduction can be seen on the ElastiCache instances that support enhanced I/O multiplexing.

When ElastiCache serves more concurrent clients, enhanced I/O multiplexing provides lower latency. Why is that?

Enhanced I/O multiplexing combines commands originating from a multitude of clients, sometimes in the thousands, into a small number of channels, which is equivalent to the number of enhanced I/O threads. This reduction in the number of channels increases access locality and consequently improves memory cache usage. For example, instead of having a read/write buffer for each client, all clients share the same channel buffer. Costly memory cache misses are reduced and the time spent on a single command is shortened. The result is faster command handling, which lowers latency and increases throughput.


With enhanced I/O multiplexing, you will benefit from increased throughput and reduced latency, without changing your applications. Enhanced I/O multiplexing is automatically available when using ElastiCache for Redis 7, in all AWS Regions, at no additional cost.

For more information, see Supported node types. To get started, create a new cluster, or upgrade an existing cluster using the ElastiCache console.

Stay tuned for more performance improvements on ElastiCache for Redis in the future.

Happy caching!

About the author

Mickey Hoter is a principal product manager on the Amazon ElastiCache team. He has 20+ years of experience in building software products – as a developer, team lead, group manager and product manager. Prior to joining AWS, Mickey worked for large companies such as SAP, Informatica and startups like Zend and ClickTale. Off work, he spends most of his time in the nature, where most of his hobbies are.

Ohad Shacham is a Senior Software Engineer with more than two decades of experience. He works for AWS, where he is currently focused on ElastiCache for Redis scaling capabilities. Ohad has a PhD in Concurrent Software Verification from the Computer Science department at the Tel-Aviv University. His expertise is Distributed Systems, Automatic Vectorization, SAT Solvers and Formal Verification technologies. Prior to AWS, Ohad was a Senior Research Scientist at Yahoo! Research, a Research Staff Member at IBM Research, a Software Engineer at Intel and an Adjunct Lecturer at the Technion, the Israeli Institute of Technology. In his free time, Ohad likes to hike with his family and to Scuba dive.