Rapping otters and automated database knob-twiddling: An obvious combination in some universe or other
OtterTune to compete with Oracle automation, but also for open source databases
A university spin-out startup has announced a private beta of an automated database tuning service which its founder claims can double the performance or halve the cost of the popular AWS Relational Database Service.
Among its marketing hype, though, is the, erm, novel approach of launching a hip-hop album of beats and screeching otters. More of that later.
Originating from a project at Carnegie Mellon Database Group, OtterTune is based on the idea you can use machine learning to identify the optimal setting for database parameter knobs, a task well beyond most developers and something even seasoned DBAs struggle with, given the number of databases on the market that they might be required to manage.
With $2.5m seed funding from Accel, OtterTune emerged from a project trying to understand how far an automated database could get without accessing queries or user data – always a pain for academics to get hold of, explained its founder and CEO, Andy Pavlo, who is an associate professor of databaseology at Carnegie Mellon University.
"We said OK, if we just use the runtime metrics that the system exposes — pages read, pages written: every system generates these things — how far could we actually get to improve it? And it turns out it, actually works: that's enough of a signal for you to be able to optimise the configuration," he said.
The university database group published a paper [PDF] on the topic in 2017, which academics paid attention to, but "nobody in the industry really saw."
But Amazon's Machine Learning Division published a blog based on the paper later that year, and that's when the idea began to gain traction, Pavlo told The Register.
"That's when everyone started emailing us and was like, 'Hey, we have that problem. We'll give you money to fly a student out to set up OtterTune for us.' But I was still [a] junior faculty member at the time, so I wasn't really in a position to run a startup or like do consulting.
"Throughout 2018, people were like, 'We want to run OtterTune, help us set it up.' So last year, we decided to take the plunge and launch in the middle of a pandemic, as one does," he said.
Initially getting started on angel funding from others in the enterprise data business, including the founders of Snowflake, Databricks, and MemSQL (now SingleStore), OtterTune has been proving itself with companies such as travel website Booking.com and French bank Societe Generale.
'There's a large market for people that just want knob tuning'
Database management systems expose knobs for buffering, caching and optimiser tuning etc, that determine the performance of the system. Pavlo argued that in modern tech, companies who run DevOps don’t have traditional DBAs, so "it's the one person who set up a database who usually gets charged [with] setting up [the] next database."
Google searches might lead them to optimisations which are sensible but have limited impact.
Even when a seasoned DBA is in charge, they might be responsible for tuning hundreds of databases each with different applications and environments, so can only tune them in the most straightforward way, and rarely revisit that, he said.
Initially launching for Amazon RDS, OtterTune can be set to target performance or cost, or a combination of both.
"We've done experiments where we can show that you take the same instance type in Amazon RDS and before and after autotune, we can get double the performance. But then we can also run on a exercise that costs half as much, and get the same performance we were getting on the more expensive machine," he said.
The paid-for subscription version of Ottertune is set to support Amazon RDS for PostgreSQL and MySQL. It plans to add Oracle and Amazon Aurora MySQL in the next few months.
Of course, Oracle already had Autonomous Database technologies, which are ahead of OtterTune, because it can see the queries and the actual data.
"For a lot of people, that's a huge privacy concern. We think that there's a there's a large market for people that just want knob tuning. We think there's a large enough market for providing this for the open source community," Pavlo said.
To coincide with the launch, Pavlo plans to do a "record drop" at Percona Live, which will feature sampled audio from screeching otters mixed over hip-hop beats. From the brief snippet leaked to The Register, we can confirm it is a unique listening experience.
"The larger vision is basically OtterTune becomes a lifestyle brand; it does databases but also there's the record label and then there's a clothing line," he said.
Presumably, branded blunts and 40 oz beer bottles will follow not long afterwards.
It's worth mentioning that the snippet released uses a sample from the The Honey Drippers' "Impeach the President", one of the most sampled beats of all time. We can only hope OtterTune's approach to database tuning is more original. ®