TL;DR
Get privacy and security 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 distributed databases, his work on Gmail and Google’s Colossus storage system, and how AI tools have increased his personal coding productivity. He shared engineering lessons including Reed-Solomon erasure coding that cut Google’s storage overhead by 33%, and career advice for founders and engineers.
Peter Mattis, co-founder and CTO of Cockroach Labs and an original creator of the open-source image editor GIMP, sat down with The Pragmatic Engineer for an interview on distributed databases, his engineering work at Google, and how AI coding tools have affected his productivity. Mattis said AI has brought him back to hands-on coding after years focused on management, and that he feels more productive than when he shipped roughly 100,000 lines of database-grade code per year before AI assistance, with no drop in quality — a personal claim that frames the episode’s broader discussion of scale, reliability, and correctness in distributed systems.
Matthi’s career spans several widely used systems. He co-created GIMP with his college roommate Spencer Kimball, and the very first version of the Google logo was made in GIMP. In 2001, Sergey Brin interviewed Mattis and made him an offer, which Mattis turned down because of the hour-long commute from San Francisco to Mountain View, according to the episode. When Google approached him again, he accepted, going on to work on Gmail and distributed storage before co-founding Cockroach Labs.
The episode includes several concrete technical details from his Google years. The first version of Gmail, which launched on 1 April 2004 with a then-astonishing 1GB of free storage, used B-trees in its storage layer: incoming messages were matched to threads via the search index, and threads and unread counts were tracked by B-trees. On Google’s next-generation file store, Mattis said the team pioneered Reed-Solomon erasure coding in a distributed file system for Colossus, the successor to the Google File System. Where GFS stored three full copies of data, Colossus stored data roughly twice while increasing redundancy — reducing file storage overhead by about 33%, according to the episode.
Matthi also described twice outperforming standard library data structures: at Google he replaced std::map (a red-black tree) with a B-tree that was faster and more memory-efficient due to spatial locality, and he later built a Swiss Table hash map for Go that was faster than Go’s built-in map — an implementation the Go team helped finish and that shipped in the Go standard library. B-trees recur throughout the conversation, from Gmail’s backend to CockroachDB’s range index, echoing the academic observation in the paper The Ubiquitous B-Tree.
Why a Veteran’s AI Productivity Claim Matters
The most newsworthy element of the interview is Mattis’s account of his own productivity with AI tools. A software veteran who shipped roughly 100K lines of database-grade code annually before AI says he is now even more productive, with no quality drop — and that AI has returned him to writing code after his role shifted toward management. This is a firsthand data point from a senior practitioner, not a vendor benchmark, in an industry-wide debate over whether AI coding assistants raise or dilute software quality. Mattis also argued that AI can multiply the impact of domain experts, a view that informs how engineering teams and hiring may evolve.
The episode also functions as a practitioner’s history of large-scale storage: Gmail’s early architecture, Colossus’s erasure coding, and CockroachDB’s design illustrate the concrete tradeoffs — latency, redundancy, and cost — that distributed database engineering still navigates today.
Top picks for "distribut databas peter"
As an affiliate, we earn on qualifying purchases.
From GIMP to Google to CockroachDB
Matthi recounted that he and Kimball nearly abandoned GIMP shortly before launch after seeing an announcement for a rival program promising everything GIMP did and more. They shipped anyway, and the competing project was never heard from again. His advice to founders: “There’s always going to be someone else working on your idea… most won’t ship it.” He also shared that he initially declined Sergey Brin’s 2001 job offer over the commute, joining another startup first that, in his words, didn’t go anywhere.
On infrastructure fundamentals, Mattis keeps “speed of light numbers” in his head, similar to turbopuffer founder Simon Eskildsen’s “napkin math.” He noted that a network round trip within a zone fell from milliseconds when Colossus was built to roughly 100 microseconds today, but that some bottlenecks are physical: because light travels fastest in a vacuum, the fastest global packet route could theoretically run up to Starlink satellites, across by laser, and back down.
“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 That Rest on One Engineer’s Account
Several figures in the episode are Mattis’s personal recollections or internal Google details that are difficult to independently verify, including the 33% storage overhead reduction from Colossus erasure coding, the internal Gmail B-tree architecture, and the ~100 microseconds intra-zone round-trip figure. His claim of higher productivity with AI — more output than his ~100K lines per year, with no quality drop — is a self-assessment rather than a measured result; he did not quantify the basis of the comparison or how quality is judged. The episode is an interview, so these statements are attributable to Mattis rather than independently confirmed.
Where the Conversation Goes From Here
The full episode, including a transcript and timestamps, is available on YouTube, Apple, and Spotify via The Pragmatic Engineer. Listeners can expect Mattis’s further discussion of the future of code review in an AI-assisted workflow and his advice on leveling up engineering skills. For readers tracking the AI productivity debate, Mattis’s account adds a senior-practitioner perspective that will likely be tested against broader industry data as AI coding tools become standard in database and infrastructure work.
Key Questions
Who is Peter Mattis?
He is co-founder and CTO of Cockroach Labs, the company behind the CockroachDB distributed database. Earlier he co-created the open-source image editor GIMP with Spencer Kimball and worked on Gmail and distributed storage at Google.
What did Peter Mattis claim about AI and his productivity?
He said that after years focused on management, AI has brought him back to writing code, and that he feels more productive than when he shipped roughly 100,000 lines of database-grade code per year pre-AI, with no drop in quality. This is a personal assessment, not a measured benchmark.
What was Colossus, and what did erasure coding change?
Colossus succeeded the Google File System. According to the episode, Mattis and the team pioneered Reed-Solomon erasure coding in a distributed file system, storing data roughly twice instead of in three full copies — reducing storage overhead by about 33% while increasing redundancy.
How were B-trees used in early Gmail?
Each incoming message was matched to a thread using the search index, and B-trees tracked threads and their unread counts in Gmail’s storage layer.
What advice did Mattis give founders?
He said to assume dozens of people have your same idea, but most won’t ship. Based on nearly abandoning GIMP before launch when a rival was announced, he advised founders to embrace competition rather than fear it.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
