What this covers

This is a one-on-one conversation between host Ryan Peterman and Mike Stonebraker, who moves from his own career to his current company. It begins with the early relational database work at Berkeley under mentor Gene Wong. Ingress was his ticket to tenure in 1976. It spread in academia because it was free software running on free Unix. It then failed commercially against Oracle Corporation, which won through sales rather than technology. Stonebraker reads that outcome as a lesson that technical merit alone does not make a business. He also describes why academic code needs a "second 90%" of hardening before it is production-ready.

The middle of the episode covers Postgres. Stonebraker explains that Ingress's fixed primitive types could not handle GIS data or a customer's bond calculation, where March 15 minus February 15 must equal 30 days. That led Postgres to its extendable type system. From there he argues that one-size-fits-all databases lose an order of magnitude to specialized systems such as column stores and vector databases. He also holds that GPUs suit indexing poorly, and that graph databases should be interface layers over relational systems. The disagreements with Google get the most room. He calls MapReduce inefficient and eventual consistency unable to keep integrity constraints like non-negative stock levels. He notes that Google later moved to transactional systems with Spanner.

The final stretch looks forward. Stonebraker describes DeBoss and the idea of OS-as-Database, citing Databricks, which moved its scheduling state into Postgres when its schedulers could not handle a million Spark jobs. He admits that venture capitalists told him he could not displace Linux, and he expects adoption to be slow. He also discusses text-to-SQL on messy enterprise warehouses, where he says current LLM approaches fail. He argues that joins across structured sources belong in SQL, not in an LLM. He closes with reading recommendations and advice on choosing work you are passionate about.

Sharpest takeaway

Stonebraker argues that specialized database systems optimized for specific workloads vastly outperform one-size-fits-all solutions, and that the future of databases lies in domain-specific architectures, intelligent text-to-SQL systems that leverage proper schema decomposition, and persistent application state managed through database technology.

  • One-size-fits-all databases sacrifice an order of magnitude in performance compared to specialized systems like column stores, stream processors, and vector databases
  • Current LLM-based text-to-SQL fails on real-world data warehouses (0% accuracy) because they lack training data, cannot handle query complexity, and struggle with messy schemas
  • Operating systems and application runtimes should replace traditional implementations with database-backed state management, as demonstrated by DeBoss architecture

The claims · ranked57 claims · weighted by value

This asset isn't compiled yet

You're seeing its claims, ranked. Compile it to build the argument threads, weight them, and check each claim against your library — the full view.

0.80

One-size-fits-all database systems are fundamentally suboptimal because specialized systems like column stores (Vertica), stream processors (StreamBase), and vector databases (Pinecone) achieve order-of-magnitude performance gains by architecting for specific workload patterns.

causalhigh valueestablishednovelty 2/4durability 4/4· Mike Stonebraker

“in each case they were an order of magnitude faster than the other guys. So it's pretty clear that one side...in with those three instances, you give up an order of magnitude uh when you're running a database system that isn't that isn't architected for your kind of stuff.”

0.80

MapReduce was fundamentally inefficient compared to distributed database systems, and Google's adoption of eventual consistency as a concurrency control mechanism was architecturally flawed because it cannot maintain integrity constraints like stock levels remaining non-negative.

causalhigh valueestablishednovelty 2/4durability 4/4· Mike Stonebraker

“Hadoop is ridiculously inefficient. And so uh at the time, you know, others, you know, Dave Dwit and others who who were involved in our 2011 paper, we understood distributed databases and understood that you could beat the heck out of Hadoop.”

0.76

GPUs are poorly suited for database operations requiring indexing because indexing does not parallelize well—searching a B-tree requires sequential pointer-following (3-4 memory accesses) that cannot be executed in parallel, whereas GPUs are optimized for SIMD (single instruction multi-data) workloads.

causalhigh valueestablishednovelty 2/4durability 4/4· Mike Stonebraker

“GPUs are you know sum simd you know single instruction multi- data and that's that's the anathema of indexing And so whenever indexing is the right answer, they're probably not a good idea.”

0.76

Databricks' schedulers from OS researchers couldn't scale to scheduling a million Spark jobs, so they solved the problem by putting all the scheduling data in a Postgres database and running a Postgres application to do the scheduling.

causalhigh valueestablishednovelty 2/4durability 4/4· Mike Stonebraker

“He said there was no we tried all the all the schedulers written by the OS folks and they they couldn't they didn't scale. So we put all the scheduling data in a Postgress database and basically a Postgress application was doing scheduling”

0.74

The commercial Ingress Corporation was founded in 1980 to move Ingress to supported platforms (VMS, DEC) and provide commercial support, becoming the pathway to commercial viability after academic prototyping had proven the relational model's validity.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“it was clear the only way out of that was to start a company. And so in 1980 uh we got venture capital as it existed then and started Ingress Corporation to move uh Ingress to uh to Dex uh VMS uh you know a real a real operating system and we had a real company that would support Ingress and that was the start of the commercial journey.”

0.74

Finding a mentor early in your career who understands the field is crucial to success; Stonebraker credits Gene Wong at Berkeley for his mentorship in the early days of relational database research.

normativehigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“you're way ahead if you get adopted by a mentor who knows the robes. So Gene Wong, who who is still alive and still still kicking, uh took me under his wing and said, 'Well, let's do something together.'”

0.74

Getting academic software to production quality requires doing the 'first 90% to get something running,' then the 'next 90% to get it working well'—the second 90% involves hardening, optimization, and reliability work that is often underestimated.

normativehigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“a lot of people would build prototypes which were sort of studenty like code, which means you could get it to run, but if you gave it to anybody else, they couldn't. So we put in the first 90% to get something we could run and then for whatever reason we put in the next 90% to get it where it really worked.”

0.74

Ingress failed as a commercial product in 1972 because it was based on a conceptually superior but commercially immature approach, whereas Oracle succeeded through aggressive sales and marketing tactics despite inferior technology, teaching Stonebraker that business success requires more than technical merit.

causalhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“I saw that Ingress was was certainly better than what they were offering, but they were still competing somehow. How how did they compete? Uh Larry Ellison is a fabulous salesman and he at the time he he would he made present tense and future tense indistinguishable and so he basically lied to customers.”

0.74

A customer of Ingress needed to calculate interest on financial bonds using a non-standard date arithmetic where March 15 minus February 15 equals 30 days (because months have fixed 30-day lengths for bond calculations), which Ingress couldn't support because date subtraction was hardcoded.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“He he said he was dealing with with bond financial instruments and for some reason I mean you got the same amount of interest on a on his financial bonds during each month no matter how long the month was. So he had the date you bought the bond, the date you sold the bond. He wanted to do a subtraction, multiply it by the coupon rate, and say that's what that's the interest we paid you. But of course, his version of subtraction was March 15 minus February 15th is 30 days because that's the definition of his calendar.”

0.73

PostgreSQL's core architectural innovation was implementing an extensible type system allowing users to define arbitrary custom data types with efficient support, enabling applications from GIS to financial bond calculations that could not be supported by fixed-type systems like Ingress.

factualhigh valueestablishednovelty 2/4durability 4/4· Mike Stonebraker

“Postgress was engineered to have an extendable type system. So you could have whatever data types you wanted and they were very efficient and that was the main gist of Postgress was that it had that flexibility.”

0.70

Multi-agent transactions involving coordinated database updates require atomic semantics where either all updates commit together or the entire transaction appears to never have happened, preventing inconsistent intermediate states when agents fail or disagree.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“if you want to write an agent or two agents that move a $100 from my account to your account. And so you debit my account, you increment your account and these two agents have to agree to commit or you have to back everything out, which is to say the workflow needs to be what I called atomic, which is it all happens or it looks like it never happened.”

0.70

Google pursued eventual consistency as a concurrency control strategy, but this approach only solves one particular kind of problem and very rarely occurs in practice.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“Google also had the opinion that eventual consistency was the right way to do concurrency control. And so that was postulated from on high by Google all during that same period of time. And it it wasn't and all the database people said, you know, you're out of your frigin mind because it doesn't it solves one particular kind of problem but only and that very rarely occurs in practice.”

0.70

When you want to talk to multiple structured databases simultaneously (e.g., a data warehouse and a CRM system), doing structured data joins using an LLM is a bad idea, and you're much better off leaving data in tables and doing joins in SQL.

normativehigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“Uh number two, the minute you want to talk to two different structured databases, you know, like your data warehouse and your CRM system, then it's pretty clear to me that doing a structured data join using an LLM is a bad idea. It's just you're much better off you leaving them as tables and doing a join in SQL.”

0.70

MapReduce was ridiculously inefficient, and Stonebraker (along with others like Dave Dewitt) understood distributed databases well enough to beat the heck out of Hadoop with a distributed database system, which was the central claim of their 2011 paper.

causalhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“Hadoop is ridiculously inefficient. And so uh at the time, you know, others, you know, Dave Dwit and others who who were involved in our 2011 paper, we understood distributed databases and understood that you could beat the heck out of Hadoop. uh with a distributed database system, which is basically what that 2011 paper says. And of course, it was it's true.”

0.69

Unix becoming the dominant operating system enabled Ingress to achieve wide academic adoption because it was a free database system that ran on the free, popular Unix platform.

causalhigh valueestablishednovelty 1/4durability 3/4· Mike Stonebraker

“over the next couple years about a hundred universities started running it because Unix became the big thing. And so this was a database a free database system that ran on Unix. And so it was quite popular in in the academic world.”

0.69

Stonebraker's tenure at Berkeley depended on publishing significant research during his first five years; the Ingress project served as his 'ticket to getting tenure,' which was achieved in 1976.

factualhigh valueestablishednovelty 1/4durability 3/4· Mike Stonebraker

“I was an assistant professor at Berkeley. As you know, if you're an assistant professor, you have to you have about you get five years to prove that you're a big and they fire you or they give you tenure. So, Ingress was my ticket to getting tenure which happened in 1976.”

0.69

Google eventually abandoned both eventual consistency and MapReduce in favor of conventional transactional systems when Jeff Dean implemented Spanner, recognizing that data correctness was necessary for their use cases.

factualhigh valueestablishednovelty 1/4durability 3/4· Mike Stonebraker

“So, we talked a million hours ago about referential integrity. So referential integrity in a sales system is uh integrity constraint is stock is greater than minus one and that fails with eventual consistency. And so uh Jeff Dean finally of Google finally figured that out and uh when they did Spanner, Spanner had a conventional transactional system and so Google comp uh completely abandoned eventual consistency and completely abandoned map reduce.”

0.68

Graph databases are almost never the performant option for their stated use cases, and instead of building specialized graph database systems, companies should build graph user interface layers on top of relational databases to provide the graph abstraction.

factualhigh valuecontestednovelty 2/4durability 3/4· Mike Stonebraker

“They're supporting a graph-based database system, and it's well understood that a graph-based database system is almost never the performant option. And so, if you want a graph, if you want, if you like the idea of having a user interface that deals with nodes and edges, that's fine. put put a layer on top of a relational database system that gives you that user model.”

0.68

VCs uniformly told Stonebraker and the DeBoss team that they were dreaming if they thought they'd displace Linux, but found the programming language features (extensions allowing database-backed state) to be compelling, leading to DeBoss Inc founding in 2023.

factualhigh valueestablishednovelty 0/4durability 4/4· Mike Stonebraker

“we talked to the VCs who to a person said you're you're dreaming if you think you're going to displace Linux. However, this programming language stuff is really nifty. We had what amounted to extensions to JavaScript that would allow any any program to have all of the nice features of a database system. You know, stuff was durable. You could have transactions. If it failed, you'd fail over. You know, it was all that nifty stuff. So we got funded uh to start a company in 2023 and that was Debboss Incorporated”

0.68

The CODASIL proposal for databases was a low-level spaghetti network architecture where you executed queries by following pointers, making it hard to debug and tightly coupled to physical storage, so that any schema change required throwing away and restarting the entire database.

factualhigh valueestablishednovelty 0/4durability 4/4· Mike Stonebraker

“the competitors were a thing called the Kotil proposal which you're probably too young to have ever heard of. And so it was a low-level spaghetti network proposal uh where you where you executed queries by following pointers and then the the alternative was the IBM proposal which was a hier thing called IMS which is still available and it's hierarchical data... The Kodasil proposal had all kinds of bad properties. Uh besides being low-level and and really hard to debug, uh it also had the property that if anything changed in your what's now called your schema, you basically had to throw away everything and do it all again because it was absolutely rooted of the physical level.”

0.68

If database-backed OS implementations are objectively superior, adoption will eventually occur, though the time constant for such paradigm shifts in computing is substantial (e.g., Java took 10 years to achieve widespread adoption).

forecasthigh valuecontestednovelty 2/4durability 3/4· Mike Stonebraker

“if it's objectively true, then maybe it will take over...I think it took Java 10 years to become widely accepted. I just think the time constant is substantial.”

0.66

The optimal strategy for career satisfaction is to find work you are genuinely passionate about, because this passion matters more than financial compensation for most people's long-term life satisfaction.

normativehigh valuecontestednovelty 1/4durability 4/4· Mike Stonebraker

“find something you're passionate about and and you will, you know, either you won't starve. You may not make a lot of money, but I think chances are you'll be happier than if you do something you're not passionate about. Because I think a lot of people I know view their job as simply a job and life is what happens between 5:00 pm and 8 am. I don't feel that way at all. I really like what I do. Wouldn't matter whether I made a lot of money or didn't.”

0.66

Query optimization was the hardest part of building the original Ingress system, and it remains the most algorithmically difficult challenge in database implementation across the industry.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“Uh query optimizer. And why was that hard? It's tough. It's it's just algorithmically difficult. It's still if you ask most any senior database programmer what's the hardest hardest part, they'll still say the optimizer.”

0.66

IBM's hierarchical IMS database was recognized as fundamentally limited even by IBM itself at the time, as they added hacky network structures to compensate for the fact that trees could not represent many real-world data relationships.

factualhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“the alternative was the IBM proposal which was a hier thing called IMS which is still available and it's hierarchical data. It's a tree...even at the time, IBM realized that trees were not general enough to to solve many people's problems. So they hacked on a a way to make it a limited network structure. So it was clear that was a horrible hack.”

0.66

Ingress could not efficiently support geographic information systems (GIS) because its fixed set of primitive data types (integers, floats, text strings) could not be extended to handle GIS types like points, lines, and polygons, which required custom indexing and operations.

causalhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“to support a GIS system, you need points, lines, polygons, line groups, that sort of stuff. And it was clear that Ingress couldn't do it because the data types we put into Ingress were the standard ones, integers, floats, uh, text, strings, and you couldn't support, you couldn't efficiently support uh, GIS types on top of that.”

0.66

Relational databases were clearly superior to existing alternatives (CODASIL's low-level pointer networks and IBM's hierarchical IMS systems) because they abstracted away physical-level implementation, allowing schema changes without rebuilding the entire database.

causalhigh valueestablishednovelty 1/4durability 4/4· Mike Stonebraker

“it was clear that was a horrible hack. The Kodasil proposal had all kinds of bad properties. Uh besides being low-level and and really hard to debug, uh it also had the property that if anything changed in your what's now called your schema, you basically had to throw away everything and do it all again because it was absolutely rooted of the physical level. Whereas Ted Cod stuff made perfect sense.”

0.64

Ingress failed in the academic market because Arizona State University could not deploy it without COBOL support on Unix, forcing them to use an unsupported operating system, unsupported database system, and the absence of COBOL left the project unviable.

causalhigh valueestablishednovelty 1/4durability 2/4· Mike Stonebraker

“Arizona State University considered running Ingress on their student records data, all 40,000 students worth. And they could get over that they had to get an unsupported operating system from Bell Labs. They could also get over they had to uh run an unsupported database system from these guys at Berkeley. But the project went down in flames when they realized there was no cobalt available for Unix. and they were a cobalt shop.”

0.64

DeBoss makes workflows faster and easier to use than competitors, with a business model focused on getting leaf-level (junior) programmers interested first, then convincing them to try the product, with recent early success among startups and beginning success with large enterprises.

factualhigh valueestablishednovelty 1/4durability 2/4· Mike Stonebraker

“So it has very very nice properties and is a great deal faster and a great deal easier to use than the competition. So the company is selling and innovating in this area... the answer is you want to make state of your application persistent when you put it in the database. uh and then it and then you figure out how to do it fast... business model as we were talking earlier is very much get and get leaf level programmers interested. So it's been very much uh you know tell us leaf level programmer tell us what you need that we don't have get it quickly and convince people to try it and we've been very very successful with other with other startups who want to choose the best thing and we're starting to be to be successful with the the big boys.”

0.60

Postgres supported abstract data types, stored procedures, and inheritance features (which were wanted by AI researchers at the time), and also supported time travel (though the implementation was poor and was eventually removed).

factualhigh valueestablishednovelty 0/4durability 4/4· Mike Stonebraker

“Uh we also Postgress also uh supported what the AI guys at the time wanted in the way of inheritance. Uh we also supported time travel. Uh but the implementation absolutely sucked and it got taken out after a while. So there were a huge number of of really nifty things at Postgress.”

0.60

In 2004, Stonebraker was working on an academic project that became StreamBase (a stream processing engine) and had ideas for column stores for data warehouses (popularized by Vertica), which were wildly different implementations that bore no resemblance to row stores.

factualhigh valueestablishednovelty 0/4durability 4/4· Mike Stonebraker

“In 2004 when I wrote the paper, we had an academic project which was building what became streambase. And so a stream processing engine looks nothing like a relational database. And we had the gist of an idea for column stores for the for data warehouses which was popularized by Vertica looks nothing like a row store.”

0.57

PostgreSQL is optimal for low-end use cases (small databases, modest workloads) but not competitive for large data warehouses because it lacks column store capabilities and multi-node support, which are table stakes for scaling to terabyte-scale data.

factualhigh valueestablishednovelty 0/4durability 2/4· Mike Stonebraker

“I think that's still true. I mean, I think Clickhouse is a column store. Pine cone is faster than userdefined types on on textbased vector processing. And so I think it's it's still very much the case and I think there's no difficulty putting a common parser on top of multiple implementations. uh Postgress has so far chosen not to do that. they don't implement a column store and so I think they are not they are not competitive you know on sizable data warehouses they also don't have multi-node support again for people with big data warehouses that's table stakes”

0.57

DeBoss is an academic project (started 2019-2020) that treats most operating system functions as data management problems and proposes replacing the upper half of Linux with database-backed implementations of scheduling, filesystem, and state management.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Mike Stonebraker

“We started the academic project in 20 19 2020 something like that...by and large most everything you do in an operating system is managing data at scale and you should do that using database technology. So why don't we just replace at least the upper half of Linux with a database system.”

0.56

Once you have established yourself, choose research areas that diverge from mainstream trends and are not 'going with the flow,' as contrarian research positioning has higher potential impact.

normativehigh valuespeaker onlynovelty 2/4durability 4/4· Mike Stonebraker

“pick some area that isn't, you know, like our our stuff, you know, which is called Rubicon, is definitely not going with the flow. So, choose something that's not that isn't going with the flow and try and make it work.”

0.55

Identifying non-intelligent individuals during interviews is straightforward through deep technical questioning: asking about their thesis, implementation details, error handling, process architecture, and design trade-offs quickly reveals gaps in understanding.

factualhigh valuecontestednovelty 1/4durability 3/4· Mike Stonebraker

“you talk to them and and it it rapidly you can rapidly surface whether they're smart or not. You know, what was your master's thesis? What did you do? Uh well, how did it how did it exactly work? Well, how did you deal with error conditions? Uh [snorts] how many processes did you have? Why didn't you use threads? I only you ask them technical deep technical quest questions.”

0.54

Large language models score 0% accuracy on real-world data warehouse text-to-SQL tasks despite 85% accuracy on public benchmarks (Spider, BIRD) because real data contains domain-specific idiosyncratic data not present in training data, has schema complexity (100+ lines of SQL vs. 10-20 lines) and messy naming conventions (underscores, non-mnemonic column names) that public benchmarks lack.

causalhigh valuespeaker onlynovelty 3/4durability 3/4· Mike Stonebraker

“on our benchmarks, uh, large language models get 0%. And if you enhance them with rag and and all the tricks goes to 10%...Data warehouse data is not in the pile. And there's an adage that if you haven't seen the data a couple times before, you have no chance of regurgitating it...query complexity on spider and bird is maybe 10 to 20 lines of SQL. Real world data warehouses it's 100 lines of SQL.”

0.53

Most agentic AI applications currently in production are read-only, meaning they generate predictions or recommendations without modifying persistent state, but the industry will inevitably shift toward read-write agents that modify data, creating massive demand for transactional database systems.

forecasthigh valuespeaker onlynovelty 3/4durability 2/4· Mike Stonebraker

“so far it's probably twothirds of the customers are doing agentic AI...so far Most of Agentic AI is read only...fairly quickly this the whole world is going to move to using you know agents to do read write applications”

0.52

A database system should only be maintained if it is performant in a large enough market to justify the ongoing maintenance burden; otherwise it should be retired even if it remains in use.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“you should retire any database system that isn't performant in in a big enough market to justify the maintenance.”

0.52

Larry Ellison employed deliberately deceptive sales practices, including misrepresenting features not yet implemented (such as referential integrity) as available, shipping non-functional software for customers to debug, and making present and future capabilities indistinguishable in presentations.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“Larry Ellison is a fabulous salesman and he at the time he he would he made present tense and future tense indistinguishable and so he basically lied to customers. He would ship stuff that didn't work and h have his initial customers help him debug it.”

0.52

The Munich Department of Transportation manages citizen complaints involving six intersecting data domains: trolley schedules (SQL), light sequencing (SQL), intersection geometry (CAD), German federal regulations (text), and Munich city regulations (text), requiring cross-domain query capability.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“we're working with the Department of Transportation in the city of Munich, Germany, and they have six people full-time who are answering citizens complaints, queries, which are of the form, how come I don't have enough time to cross this intersection next to my house...Their database is the trolley schedule, that's SQL. The light sequencing, that's SQL. The intersections, that's CAD. the federal you know country of Germany regulations of this stuff that's text city of Munich regulations for this stuff which is text”

0.52

Operating system designers and programming language researchers initially reacted to the DeBoss proposal with threatened defensiveness, viewing it as territorial encroachment by database researchers rather than evaluating it on technical merits.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“back in the academic project when I'd mentioned that to operating system folks they would get very very threatened which is this is the database guys trying to take over their turf and I think the programming language guys ditto”

0.52

Following your passion is necessary but not sufficient for career success; financial viability must also be achieved, yet most people and society tell young people to follow passion anyway because the alternative advice is socially unacceptable.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“Both my wife and I said, 'Follow your passion. Somehow the money will work out.' And I don't believe that for a minute, but I think that's what you have to tell your kids.”

0.52

DeBoss Incorporated's core value proposition is providing application language extensions (in TypeScript, Java, Go, Python) that implement transactional durability, automatic failover, and atomic workflow semantics, making applications database-backed without requiring developers to explicitly manage state or transactions.

factualhigh valuespeaker onlynovelty 3/4durability 3/4· Mike Stonebraker

“the workflow that that deboss supports in [snorts] those four languages is the steps in in a workflow, the individual micro apps, whatever you want to call them, are transactional. Uh workflows are durable. So that once you do a step it's not forgotten.”

0.52

Stonebraker both says 'follow your passion, somehow the money will work out' and immediately doubts it, acknowledging the tension between the genuine advice he would give and what he tells his family to believe.

factualhigh valuespeaker onlynovelty 1/4durability 4/4· Mike Stonebraker

“Both my wife and I said, "F follow your passion. Somehow the money will work out." And I don't believe that for a minute, but I think that's what you have to tell your kids.”

0.49

Amazon is supporting approximately 15 different database systems, of which only about 3 are justified; the remainder (including graph databases) are redundant because specialized functionality can be implemented as layers on top of relational systems rather than requiring separate database engines.

normativehigh valuespeaker onlynovelty 2/4durability 2/4· Mike Stonebraker

“Amazon's problem is that they are supporting you know 15 different database systems and that's about 12 too many...they're supporting a graph-based database system, and it's well understood that a graph-based database system is almost never the performant option. And so, if you want a graph, if you want, if you like the idea of having a user interface that deals with nodes and edges, that's fine. put put a layer on top of a relational database system that gives you that user model.”

0.48

Stonebraker's wife had both undergraduate and master's degrees in computer science but wanted to be a K-12 teacher; her parents prevented this career path due to financial concerns, and she has regretted that decision for the rest of her life because she was never passionate about computer science.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Mike Stonebraker

“My wife is is a good example. So she has an she has a master's degree in computer science, undergraduate degree in computer science, and she wanted to be a teacher, you know, you know, K K12 teacher. And her parents said, 'You can't do that. It doesn't pay enough money.' And so I think and I think she ever since that time has regretted that decision. She wasn't passionate about doing computer science. it was simply a trade.”

0.48

The right career strategy if you're starting out today is to 'think outside the box' and 'think crazy thoughts' and attempt to execute them, despite their apparent riskiness, because the learning process of attempting ambitious goals teaches more than conservative choices.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Mike Stonebraker

“So starting off doing something that was that crazy was really pretty crazy. And and you know you you effort and you make stuff work and you learn along the way. And so I think the answer is think outside the box. Think crazy thoughts and try and do them.”

0.48

Stonebraker prefers academia over large industry positions because industry roles impose management hierarchies that limit autonomy, restrict ability to publish and speak at conferences, and prevent critique of competitors' approaches.

normativehigh valuespeaker onlynovelty 1/4durability 3/4· Mike Stonebraker

“that gives you a boss...And that gives you company rules, limits your ability to publish, limits your ability to go talk at conferences, uh limits your abil your ability to go go poke at what what various competitors are doing that they won't tell their competitors.”

0.47

Human SQL programmers with knowledge of the schema achieve very high accuracy (90% or better) on the Beaver text-to-SQL benchmark, establishing a human baseline that LLMs fail to approach.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“So once you disambiguate the the text, a a knowledgeable SQL user programmer with the schema will do will will get very high accuracy...Okay. Like 90% or something at least.”

0.47

A file system implemented on top of a database management system performs faster than the native Linux file system, and database-backed scheduling engines are competitive with OS schedulers while providing high availability failover without additional engineering.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“A file system written on top of a DBMS is faster than than the Linux file system. The scheduling engine is competitive with other scheduling engines. You can make everything fail over. So, you get high availability without having to do anything else.”

0.47

For text-to-SQL systems to be viable, they must provide the retrieval system with complete context (FROM clause and all JOIN terms) and break complex queries into simpler pieces, rather than expecting LLMs to reconstruct this information from schema alone.

normativehigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“if you don't have all the join terms and you don't have the from clause, you're toast. What's more, if you don't break down the query into simpler pieces, you're toast. So that says to me that uh you want to give your retrieval system simpler pieces which include the from clause and include join terms.”

0.47

Stonebraker published a benchmark called Beaver, which is an anonymized and abstracted version of real production data warehouse schemas from four organizations, designed to provide a more realistic test of text-to-SQL systems than public benchmarks like Spider and Bird.

factualhigh valuespeaker onlynovelty 2/4durability 3/4· Mike Stonebraker

“So we've been So what do you do? So well first of all we published our benchmark. It's a thing called Beaver, which is an anonymized and abstracted version of these four data warehouses. And so if you think you're really good at doing text to SQL, try a real benchmark, not a fake one.”

0.45

Clickhouse is a column store, and Pinecone is faster than user-defined types on text-based vector processing, demonstrating that specialization in database architecture continues to be true today as it was in 2004.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Mike Stonebraker

“I think Clickhouse is a column store. Pine cone is faster than userdefined types on on textbased vector processing. And so I think it's it's still very much the case”

0.45

Computer science may not be a growth industry going forward, and health care and building trades are safer career bets for 18-year-olds than majoring in computer science.

forecasthigh valuespeaker onlynovelty 1/4durability 2/4· Mike Stonebraker

“computer science may well not be a growth industry going forward. And I'm not sure I would recommend 18 year olds to major in computer science. I mean, I think health health care and and the building trades are are are safe bets and everything else looks much riskier.”

0.43

Identifying extremely talented software engineers is straightforward: they accomplish approximately 3x the output in school that would normally be considered reasonable for the effort expended.

factualhigh valuespeaker onlynovelty 1/4durability 3/4· Mike Stonebraker

“I have a good feel for how difficult stuff is. if they get 3x the am the the amount done, you know, in school that I think is reasonable, then then they're incredible.”

0.41

After Postgres was acquired by InformX, Stonebraker worked part-time for this 2,000-person company but didn't feel he could make a difference because it was bureaucratic and whatever the president wanted, he got.

factualhigh valuespeaker onlynovelty 0/4durability 3/4· Mike Stonebraker

“after the commercial version of Postgress got acquired by InformX. You know I was working part-time for InformX which was a 2,000 person company and I didn't feel like I could make a difference because it was bureaucratic and and whatever the president wanted he got.”

0.23

Stonebraker has difficulty interacting with people he perceives as unintelligent and cannot stand such interactions, viewing this as a personal limitation.

factualspeaker onlynovelty 0/4durability 3/4· Mike Stonebraker

“I have a hard time interacting with people I think are dumb and that again. So, I guess I I have I have some problems with with big companies.”

0.20

Many people view their job as purely transactional income, with 'life' being what happens outside work hours (5pm-8am), whereas Stonebraker has organized his entire life around work he finds meaningful and wouldn't change even if money weren't a factor.

factualspeaker onlynovelty 0/4durability 3/4· Mike Stonebraker

“Because I think a lot of people I know view their job as simply a job and life is what happens between 5:00 pm and 8 am. I don't feel that way at all. I really like what I do. Wouldn't matter whether I made a lot of money or didn't.”