Future Tech

Redis justifies open source shift with fresh hardware, LLM cost-saving features

Tan KW
Publish date: Thu, 12 Sep 2024, 10:57 PM
Tan KW
0 476,810
Future Tech

Interview Redis is the most popular database on AWS, which is, of course, the most popular cloud. The fact the relatively little known database, which launched in 2009, punches above its weight against well-established rivals might owe a lot to its reputation as a handy off-the-shelf cache developers know and love. Yet for the last couple of years, it has been champing at the bit to be much more.

The in-memory database vendor, as readers will recall, switched to a dual-license approach, imposing more restrictive terms, earlier this year. The company now says this has allowed it to introduce more innovative features, which it says promise an 80 percent cost reduction in call memory, and slicker more efficient management for GenAI.

In March, Redis - the company formerly known as Redis Labs - announced that from 7.4, Redis would be dual-licensed under the Redis Source Available License (RSALv2) and Server Side Public License (SSPLv1), whereas it had previously come under the BSD 3-clause license, a more permissive arrangement which allows developers to make commercial use of the code without paying, a stance which has been the subject of some debate.

The decision followed a string of database vendors opting for so-called source available licenses to mitigate the risk of third parties - specifically cloud providers - using the open source code to provide rival database services.

Speaking to The Register after a volley of technology updates, Redis CEO Rowan Trollope says switching to a more restrictive approach to licensing had been key to improvements in the database.

"As a result of changing the license, we were able to bring all of the technology that we had been working on a separate distribution, and combine them into the new licensed, free software Community Edition. That's a big benefit, for developers," he tell us.

Trollope explains that the SSPL license only really "applies to Amazon and Google." He says if a third party wants to deliver the Redis open source as part of a cloud service offering that is directly competitive with Redis, they must publish all of the source code or get a commercial agreement with Redis Incorporated.

He confirms Microsoft has agreed commercial terms with Redis while Amazon and Google have not. Meanwhile, the Redis license is more permissive than other companies which followed a similar path, including Elasticsearch, MongoDB, MariaDB, and HashiCorp, he adds.

In April, cloud giants AWS, Google, and Oracle responded to Redis's shift in licensing, joining with the Linux Foundation to create an open source fork of Redis called Valkey. Within weeks it had a release candidate and added more backers including Alibaba Cloud, Huawei, and telecoms provider Verizon.

Trollope says Redis had expected such a move.

"We anticipated that they would fork because that's exactly what Amazon did with Elasticsearch; they forked it. We think that it's much better for us to be able to innovate freely on our code base without fear of code being taken by cloud service providers and resold as part of their offers," he says.

Not everyone agrees. Some developers have opined that the paid-for tier "Enterprise Edition" approach from database companies creates a sales-oriented approach which sees them "squeezed by the vendor." Others complain that the open source approach is used by some companies as a teaser to get developers interested "and then when investors want revenue growth, the rug gets pulled."

The Elasticsearch story has another perspective. On Hacker News, AWS VP Adrian Cockcroft, said the core of the issue was "that AWS wanted to contribute security features to the open source project and Elastic wanted to keep security as an enterprise feature, so rejected all the approaches AWS made at the time." He said Elasticsearch had wanted to control the project and slow contributions that reduced its differentiation.

Nonetheless, Redis is claiming to offer significant cost/performance advantages with new features relying on SSD, which it first revealed to The Register last year. Redis is designed as an in-memory database, but Redis Flex is designed to run on DRAM and SSDs, which the company claims makes it faster and cheaper than other memory solutions like ElastiCache, Memorystore. It offers an 80 percent hardware cost reduction compared with Redis Software, Redis Cloud, and self-managed Redis, without needing to change your existing data infrastructure, the company said.

"This going to really help the people that are using the free software where this capability doesn't exist, because essentially, you can save 80 percent of your hardware costs by upgrading to Redis Flex," Trollope tell us. A free version of Redis Flex would be available without all the features of the paid-for version, he adds.

Another development Redis is promoting with its new approach to licensing are features to support the application of GenAI in business, including Semantic Caching, which is designed to make more economical user of LLM queries.

According to Redis, Semantic Caching "interprets and stores the semantic meaning of user queries, allowing systems to retrieve information based on intent, not just literal matches."

Trollope says: “One of the big problems we dealt with our large enterprise customers is when they go to put chatbots into production, they immediately start to look at the bills from OpenAI and go, 'Wait a second, this is too expensive.' So they want to be able to cache the questions, but you can't do a straight text caching like you would, like you would sort of do before Gen AI. You really want to take the question, convert the question into a vector embedding and store a potential answer from the LLM for that question. [What] that allows you to do is to match similar questions in the vector space with a single answer, and what we found is, in typical consumer chatbot scenarios with our enterprise customers, that's reducing the calls to the LLM by about 30 percent, making a saving on your OpenAI API bill.”

The Redis CEO argues that the new licensing stance made the features possible. “We were able to take years of great tech that we weren't able to put into the open source product because it would have been taken by the CSPs without any remuneration. We were able to take all that technology and put it in there.”

Whether the company can continue to do that without killing the golden egg laid by open source community goose will continue to be debated. ®

 

https://www.theregister.com//2024/09/12/redis_justifies_open_source_shift/

Discussions
Be the first to like this. Showing 0 of 0 comments

Post a Comment