Sizzling SQL databases

Review: DeepSQL outruns Amazon Aurora

DeepSQL makes index-heavy ingestion and queries go really fast, outperforming Aurora by twofold in our benchmarks

At a Glance
  • Deep Information Sciences DeepSQL 5.6.x

Editor's Choice

When I reviewed Amazon Aurora last October, I noted that its level of performance was far beyond any I had seen from other open source SQL databases, and it did so at a far lower cost than you would pay for an Oracle database of similar power. Considering that Aurora is a drop-in replacement for the ubiquitous MySQL, Amazon certainly had a winner on its hands.

In an investor call late in 2015, Phil Hardin, director of investor relations at, said the Aurora relational database engine was a technological high point for AWS, and it was “our fastest-growing service ever.” As we’ve long known in tech, pioneers usually wind up with arrows in their backs.

The arrow in Aurora’s back I’ve now evaluated is DeepSQL, another drop-in replacement for MySQL. Deep Information Sciences claims that DeepSQL is faster than Aurora by multiples, depending on the task, and I challenged them to prove it.

How DeepSQL works

DeepSQL uses data structures that are a departure from the b-trees (more or less read-optimized) and log structured merge trees (more or less write-optimized) of classic (circa 1970s) databases. Taking advantage of the evolution in computer architecture over the years (more and faster RAM, more and faster CPUs, faster disks), DeepSQL constantly tries to reach the theoretical minimum disk seek costs: writes should have a seek cost of 0, and reads should have a seek cost of 1. To write without seeking, you can only append to the data log file, then append to the index log files. To read with anything close to one seek, you need to keep indexes and caches and summarizations in memory.

To continue reading this article register now

How to choose a low-code development platform