TL;DR
Get networking and server gear delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
Peter Mattis, co-founder and CTO of Cockroach Labs and an original creator of GIMP, appeared on The Pragmatic Engineer podcast to discuss his career spanning GIMP, Gmail, Google’s Colossus storage system, and distributed databases. He shared engineering lessons on B-trees, erasure coding, and why he believes AI tools have made him more productive without a drop in quality.
Peter Mattis, co-founder and CTO of Cockroach Labs and an original creator of the open-source image editor GIMP, appeared on The Pragmatic Engineer podcast to discuss his career in distributed systems — from the first Google logo, drawn in GIMP, to Gmail’s storage layer and Google’s Colossus file system — and why he believes AI coding tools have made him more productive, not less careful. The episode, published with a full transcript, covers how large-scale databases stay fast, reliable, and correct, and includes Mattis’s reflections on engineering craft built over more than two decades.
Mat tis’s path to distributed databases began in open source. He and his college roommate Spencer Kimball created GIMP, and the first version of the Google logo was made using the tool, according to the report. In 2001, Sergey Brin reached out to Mattis with a job offer — which Mattis initially turned down because he did not want an hour-long commute from San Francisco to Mountain View. After a stint at another startup that, in his words, “didn’t go anywhere,” he accepted when Google came back a second time. At Google he worked on Gmail and distributed storage.
Several concrete engineering details from the episode stand out. The first version of Gmail, launched on 1 April 2004 with 1GB of free storage — an offer so generous many assumed it was an April Fool’s joke — used B-trees in its storage layer. Each incoming message was matched to a thread using the search index, and threads and their unread counts were tracked by B-trees. Later, Mattis and his team pioneered Reed–Solomon erasure coding in Google’s distributed file system for Colossus, the successor to the Google File System. Whereas GFS stored three full copies of data, Colossus stored data twice while increasing redundancy — and, per the report, reduced Google’s file storage overhead by 33%.
Mat tis also described twice beating standard library data structures in performance. At Google, he noticed std::map appearing in memory profiles, identified its red-black tree implementation as the cause, and built a B-tree with nearly the same semantics that was both faster — due to spatial locality — and smaller, because it used fewer pointers. He later did something similar with Go’s map, building a Swiss Table implementation that was faster; that work, with help from the Go team, eventually landed in the Go standard library. The report notes that B-trees recur across his work — Gmail’s backend, the std::map replacement, and CockroachDB’s range index — a phenomenon covered in the paper “The Ubiquitous B-Tree.”
Why a Storage Veteran’s Lessons Matter Now
The episode lands at a moment when infrastructure teams are rebuilding storage and search layers from scratch, and when debates about AI’s effect on engineering quality are intense. Mattis — described in the report as a veteran who regularly shipped roughly 100K lines of database-grade code to production each year before AI tools existed — says he is even more productive today with no drop in quality, and that AI brought him back to writing code after years focused on management. That claim, from someone with deep systems credentials, is a data point in the ongoing argument over whether AI assistance helps or harms code quality.
His technical observations also translate directly to current infrastructure practice. Erasure coding, which he helped pioneer at Google scale, is now standard in cloud storage systems. His point that a network round trip within a zone has fallen from milliseconds when Colossus was built to roughly 100 microseconds today — about 10x faster — illustrates how latency budgets shift as hardware improves, sometimes leaving the speed of light itself as the bottleneck. He argues the fastest global packet route may eventually run through space: up to satellites like Starlink, across by laser, and back down.
Top picks for "distribut databas peter"
As an affiliate, we earn on qualifying purchases.
From GIMP to CockroachDB
Mat tis and Kimball developed GIMP as students, and nearly shelved it weeks before launch after seeing an announcement for a competing program that promised to do everything GIMP did and more. They shipped anyway, and the competing project was never heard from again. Mattis drew a founder’s lesson from the episode: “There’s always going to be someone else working on your idea… assume that dozens of people have the same idea you have, but most won’t ship it.”
After Google, Mattis co-founded Cockroach Labs, the company behind CockroachDB, a distributed SQL database designed for resilience and horizontal scale. The podcast conversation connects that work back to earlier systems: Gmail’s early storage challenges, Colossus’s erasure-coded architecture, and his hand-tuned data structures all inform how CockroachDB handles distribution and consistency.
“There’s always going to be someone else working on your idea. You can’t get dissuaded if they pre-announce it. To founders: assume that dozens of people have the same idea you have, but most won’t ship it!”
— Peter Mattis, co-founder and CTO of Cockroach Labs
Claims Worth Noting Carefully
Several points in the episode are personal recollections or single-source claims rather than independently verified findings. The 33% storage overhead reduction attributed to Colossus’s erasure coding comes from Mattis’s account in the report, not a published Google paper cited in the material. Likewise, his assertion that AI tools improve code quality is his own experience and opinion — he presents no benchmarks, and the report offers no independent measurement of the code he ships with AI assistance. The claim that a satellite route would be the fastest global network path is presented as a reasoned argument based on light-speed physics, referencing the paper ‘Delay is not an option: low latency routing in space,’ rather than a deployed production route.
Where to Follow the Conversation
The full episode is available on YouTube, Apple, and Spotify, with a transcript and timestamps published alongside the report at The Pragmatic Engineer. Readers interested in the technical threads Mattis raises can look at the paper ‘The Ubiquitous B-Tree,’ the Starlink routing research he references, and Cockroach Labs’ ongoing engineering blog for how these distributed-database principles are applied in current products. Mattis also offers advice in the episode on leveling up engineering skills and rethinking code review in an AI-assisted workflow — topics likely to recur as more senior engineers adopt AI tooling.
Key Questions
Who is Peter Mattis?
He is co-founder and CTO of Cockroach Labs, the company behind the CockroachDB distributed SQL database. Earlier, he co-created the open-source image editor GIMP with Spencer Kimball and worked at Google on Gmail and distributed storage.
What did Peter Mattis work on at Google?
He worked on Gmail — whose first version used B-trees to track email threads — and on distributed storage, including pioneering Reed–Solomon erasure coding for Colossus, the successor to the Google File System.
What did he say about AI and programming productivity?
He said AI tools brought him back to writing code after years in management, and that he feels more productive than before with no drop in quality. This is his personal assessment based on his own experience, not a measured study.
How did Colossus reduce storage overhead at Google?
According to Mattis’s account, GFS stored three full copies of data; Colossus used erasure coding so data was stored roughly twice while redundancy actually increased, cutting file storage overhead by about 33%.
Why does he call the B-tree ‘ubiquitous’?
B-trees appear throughout his work — Gmail’s thread tracking, his replacement for C++’s std::map, and CockroachDB’s range index — and the phenomenon is documented in the paper ‘The Ubiquitous B-Tree.’
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
