apavlo

Andy from OtterTune here.

I suppose that it is possible to do something like this. We've explored the idea of synthesizing a workload that behaves like an application's real workload. The idea was to do offline testing with the synthesized workload to find optimal configurations that can then be applied to the production database. Our results were inconclusive (not enough real workloads to verify). But the idea could be extended for what you are proposing.

But in terms of relational vs graph, my opinions about the matter are public:

https://www.theregister.com/Debates/2023/03/06/great_graph_d...

It should be `item.added_rev_id = 3`. I fixed the typo. Not sure how it got mangled. Thanks!

> disregarding Recommended Music for this Paper: Dr. Dre – High Powered (featuring RBX))

Why?

> that paper seems too short to adequately explore the topic.

The paper was published in CIDR (https://www.cidrdb.org). The paper submissions for this conference are meant to be short (typically 6-7 pages) to ensure that people can get their ideas out quickly.

Is the list not showing up for you? Can you email me a screenshot: pavlo@ottertune.com

It should look like this: https://i.imgur.com/abB2NJX.png

OtterTune | REMOTE | Full-time | https://ottertune.com

OtterTune is a database optimization-as-a-service company based on research out of Carnegie Mellon University. We use ML to automatically improve the performance of MySQL/PostgreSQL databases running on Amazon. We are a remote-first company. We are backed by VC's at IntelCap, RaceCap, Accel and the co-founders of Snowflake, DataBricks, SingleStore, CockroachDB, and Duolingo.

* Front-end Engineer: https://jobs.lever.co/ottertune/7ce5d9b0-c54f-41e1-9367-b7b3...

Legit compensation + equity. 401k w/ matching. Unlimited PTO + sickdays + paid FMLA.

We are also hiring for other roles: https://ottertune.com/careers/

Email me if you have questions: pavlo@ottertune.com

> Of course at the end of the article they offer a solution - their product (and of course it’s AI enhanced)

We have been working on automatic database optimization using AI/ML for a decade at Carnegie Mellon University [1][2]. This is not a gimmick. Furthermore, as you can see from the many comments here, the problem is not overhyped.

[1] https://db.cs.cmu.edu/projects/ottertune/

[2] https://db.cs.cmu.edu/projects/noisepage/

No. It would be a major surgery on the internals. See the article for my comment at the attempt to do this with the Zheap project:

https://wiki.postgresql.org/wiki/Zheap

Legacy reasons. The idea was that you wouldn't need a WAL because the table itself is the log. And then you could support time-travel queries if you never cleaned up the expired tuples.

It's one of those things where you know it when you see it. I got 18 year old CMU students showing up knowing how write x86 SIMD intrinsics or ones with LLVM JIT experience. Dirty.

Not for the Advanced class. We cover state-of-the-art systems. Textbooks are about 10 years behind.

That class just uses my slides. It says so at the bottom of the page:

> The lecture slides used in the course are taken from Prof. Andy Pavlo's CMU 15-721 course

Just go to the source: https://15721.courses.cs.cmu.edu/spring2023/

OtterTune | REMOTE | Full-time | https://ottertune.com

OtterTune is a database optimization-as-a-service company based on research out of Carnegie Mellon University. We use ML to automatically improve the performance of MySQL/PostgreSQL databases running on Amazon. We are a remote-first company. We are backed by VC's at IntelCap, RaceCap, Accel and the co-founders of Snowflake, DataBricks, SingleStore, CockroachDB, and Duolingo.

* Marketing Lead: https://jobs.lever.co/ottertune/0501b20a-2ac9-4772-ae23-5f97...

Legit compensation + equity. 401k w/ matching. Unlimited PTO + sickdays + paid FMLA.

We are also hiring for other roles: https://ottertune.com/careers/

Email me if you have questions: pavlo@ottertune.com

> The primary con of academia is almost certainly having to live far from family.

It depends on your family.

> The man writes small databases for fun over a weekend, I mean, come on.

I am a student of Mike. He is a great man, but he would not call himself a programmer.

I spent some time looking at public documentation and FAA job listings to try to find out what DBMS they are using for NOTAM. I couldn't find anything other than some CFPs about them using Oracle and Sybase.

> but maybe what results is a relational database+ type thingy

This is called already called "object-relational" model. It was invented by Postgres in the 1980s. The relational model / SQL absorbs the best part of alternative systems and get better over time. SQL:2023 is adding support for graph queries (SQL/PCG).

Graph DBMSs are a passing fade.

You don't need to be a student to access Gradescope:

https://15445.courses.cs.cmu.edu/fall2022/faq.html#q7

It's maintained by me (also the author of this article).

We want to write a paper on the vast extensions ecosystem for Postgres at some point. It's on my TODO list for 2023.

Having so many plug-ins/add-ons available is unique to Postgres and due to Stonebraker's original vision of including extensibility as a first-class design principle in the DBMS.

I saw your engine a week or so ago. Good stuff. I wanted to add it to dbdb.io but I couldn't find an SVG version of this:

https://github.com/marsupialtail/quokka/blob/master/docs/doc...

Can you email it to me?

> But Musk got started by confounding x.com/PayPal, which does have very similar engineering needs as Twitter.

But PayPal was running Oracle on IBM bigiron machines even after Musk left them.

This is pretty much how all DBMSs work right now. You have to tell it how much memory it can use for its internal buffer pool.

I taught Judy Arrays in our Advanced DB course in 2020:

https://youtu.be/N6rhECUjdaI?t=3043

AFAIK nobody uses them because of the (unfounded) patent fears.

ART indexes are a modern implementation of radix trees / tries that is better.

> How can someone get started writing a database these days?

Take my class. It starts next week: https://15445.courses.cs.cmu.edu/fall2022/

Why would you link to an interview with unknown developers instead of a Github repo or something that describes the system architecture? AFAIK MorselsDB is a closed-source fork of Clickhouse and the only info about it are these interviews. What are you trying to achieve with this?

You would think a marketing company would be better at marketing.

Everyone should watch one of Martin's last public talks in 2021 on Youtube:

MonetDB: Scale Up Before You Scale Out

https://youtu.be/vjWRE0UnJDQ

https://db.cs.cmu.edu/events/vaccination-2021-monetdb-scale-...

But it's just a wrapper around SQLite. Skip the middleman and just use SQLite.

Postgres is not every system. Just because it requires manually refreshing matviews doesn't mean that every DBMS works that way.

Seriously, read a textbook:

http://db.cs.duke.edu/papers/fntdb12-ChirkovaYang-mat_views....

From the documentation:

> Yet SQL does not provide a way of indexing queries that involve a JOIN. Social apps that want to query "what are all the posts of all the people I follow ordered in time" must design their own systems because SQL cannot index that query (SQL will need to load all of the posts of all the people you follow over all time, and sort them in memory!)

rough query but you get the idea

  CREATE MATERIALIZED VIEW follower_posts
    SELECT f.follower_id, p.post_id, p.author_id, p.created
      FROM followers AS f JOIN posts AS p
        ON f.followee_id = p.author_id;
  CREATE INDEX ON follower_posts (follower_id, created DESC);

Please read a textbook about databases before building your own.

Make a logo and email it to me. I will add it to dbdb.io