I have had a couple of public talks from my time in Tantan, where we after a couple of years had a very large database cluster (100+ beefy servers) of Postgres running. Tantan was basically built on Postgres and scaled a lot during the time I were there, resulting in some interesting challenges. If you are just causally browsing I suggest taking a look at the second talk, "Using PostgreSQL in Tantan - From 0 to 350bn rows in 2 years" which I think is the one most generally interesting.

(Ab)using 4D indexing in PostGIS 2.2 with Postgres 9.5 to give you the perfect match

2016 March 19 at FOSSAsia 2016 / pdDay Asia 2016 in Singapore

Slide here, also online at YouTube

My Reflection: I have not dared to watch the recording fully, but if/when I have another talk I should! I'm sure there are many details to improve on. More fun and focused on a specific niche topic compared to the other talks since I wanted to do something else. I suspect it was a too niche and difficult topic, but at least the conference and Singapore was very nice!

Talk Abstract: In this talk I will talk about doing KNN (nearest neighbor) search in Postgres with PostGIS and ordering the result based on more than just physical distance. This ordering is the core of the dating app Tantan, which I will use to illustrate the talk. At Tantan we have more than one million users every day that look through 200 million potential matches each day, and we need to optimize what we show to maximize the potential chance for a match. We do that by ordering the potential matches based on a number of criteria, such as distance from you, recent activity and popularity.

Using PostgreSQL in Tantan - From 0 to 350bn rows in 2 years

2016 December 2 at PGConf Asia 2016 in Tokyo

Slides here

My Reflection: The talk I would find the most interesting to listen to myself.

Talk Abstract: In this talk I will cover how PostgreSQL together with PostGIS forms the backbone of Tantan; the workload we have, the hardware we use, the software we use and also challenges we meet and a quick look at our custom sharding solution. In previous talks I have given about how we use PostgreSQL in Tantan people have asked for a more broad view and this is what I hope to give in this talk.

Location based dating in China - 0 to 100000000 daily swipes in a year

2015 October 28th at PostgreSQL Conference Europe 2015 in Vienna

Slides here

My Reflection: This was my first major public talk! Overall I think it went very well, but my wife will forever tease me about "My name _was_ Victor..." that I somehow said in the beginning when introducing myself!

Talk Abstract: In this presentation I will talk about how we have scaled horizontally with a custom, but simple sharded solution built on schemas to handle our current and future load. About how we have come a long way with a simple strategy, using only PostgreSQL and application sharding. I bring up some optimizations we did and some things to watch out for.

Each day we store and process 100 million swipes, create 500 thousand matches between our users and respond to 10 million geospatial queries. In the app we show nearby users ranked by closeness, activity and profile information, and only those you have not seen before ranked for optimal match chance. To calculate the list of nearby available users we have a fairly complex PL/pgSQL function utilizing a gist index for the k-nearest neighbor ordering. It returns a result in around 100-200ms.