i reached out to the author and he said he used only free tiers for this test. Hence the reason there is no planetscale replication as its not available in the free world.
@Cookie note that all setups were provided by the database vendors themselves! (author here). This is not a benchmark where we set up other vendors, this is organized by Vercel and each vendor set up their own systems.
18:11 I looked at the docs and Turso is essentially a read only replica as well. The difference is for writes, you still use the edge as an endpoint but it sends the write over to the primary.
Not sure how planet scale works but I guess the benefit or turso is that you can just write to the replica and call it a day. You can offload the slow request to the primary to turso. I’d be interested in how they keep their system fault tolerant so that they don’t drop writes if a node fails
Brazil mentioned! Yeahhh, come to Brazil Prime, we will give you some Pão de Queijo and Coxinha with caldo de cana. During the night time you will drink Cachaça💚💛💙
This is aewsome! As a software engenieer based in Brazil, we plan our applications to make as few requests as possible to anything in America and in many ocasions the latency is a deal breaker. Truely life changer lol
@Tunema hey - author of the blog here. I was actually born and raised in Brazil, although by now my Portuguese sucks a bit! Glad to see you liked the website
Eager to see Prime turn back into his primary for of 'cat' (for context, the title at the time of watching is "Benchmarking Databases on Vercel | Prime Recats", in case Prime fixes it later)
The idea of having read replicas everywhere sounds great in theory, but replication lag can be a killer. Especially for those cases where you require your data be fresh everywhere and have to enforce sync writes (and sync replication). If your data can be a bit stale though, and use async replication, the more replicas the merrier.
I question when this actually is important. Stock tickers? Your avg CRUD SaaS probably won't care even for up to a minute lag. >5min and we have problems tho for sure.
Turso is interesting, they use raft consensus as far as I can tell to synchronize the various read state replicas across edges. On one of my previous projects, we combined raft with CRDTs to build a time series database and let me tell you, I wish this thing was not proprietary because I would absolutely be pushing it as the best distributed time series database out there. It just makes sense to do time series this way since you rarely are deleting or mutating data in a time series table. Anyway, I also wish that turso was open source.
The benchmark should have been done with Planetscale's read replication turned on for it to be a fair comparison. I'd be very interested to see how that would affect the results.
yeah, the 1v1 results show a big difference as is but the read replication is a pay for feature only. i reached out to the author and he said he used only free tiers for this test.
Author here! The tool is provided by Vercel, not us, and each database contributed their own setups - including Planetscale. That said, this is not really a benchmark against Planestcale - they actually did fantastic ! You are right that with read replicas planetscale (and others) can achieve this, but there is a reason they - and others - didn't add replicas in this test. It's quite expensive! The advantage of Turso, that I tried to touch on the article, is how easy and cheap (the free tier alone already has 3 locations) is to set up replication. Great comment, though! ❤
This is interesting because of the freemium perspective, but it doesn't bring fundamental shifts. I'm more interested in seeing how Cloudflare would do with their Distributed SQLite approach if they really get to make read/write replicas. At this moment, No-SQL alternatives like FaunaDB or CoakroachDB seems to be the ones actually breaking the molds with respect of edge data.
Why shouldn't you deploy at the edge ? Just read the pretty inconvenient cons on Vercel's own page ( + ignoring the fact that you have to write JS on the backend 😅)
Why don't you like C++? I know Rust is great and doesn't have hidden execution paths, but at least in C++, you can recover from Out of Memory errors. (Just to note, I really like Rust.)
C++‘s share Pointers are thread safe by default meaning they have extra over head on construction because they are atomic. Rust has Rc and Arc for specific use case. Rust io and string manipulation is leagues ahead of C++. The compiler, build tools, packages and ecosystem are better. CMake is horrible, i build recursive globing smart CMake scripts with for loops macros function… it’s horrible. Cargo is a dream to work with. Again the dependency management is just better. And Rust is just plain more fun to write in
Other than terrible ecosystem/toolchain, C++ is extremely hard to learn and extremely easy to cause some serious issues in without even realizing it. It's super feature-bloated and those features are not easily discoverable, a lot of the more modern changes are good but at this point it's almost impossible to know what's best practice, what you should do and what you should avoid, reading someone else's code can be hell even if you're already very proficient at it. There's less and less junior C++ jobs nowadays, because juniors are useless at writing C++, you're either very good at it or you're useless, or even potentially detrimental to the project if your mistakes go unnoticed. Since better alternatives exist nowadays, there's simply no reason to use it, and no reason for new devs to learn it.
won't that be very expensive? You literally have a database per edge + cross region costs are pretty high in the cloud so that will add another hidden cost just to get a couple of milleseconds.
Pretty misleading messages on their landung page. They put the text "Open source" at the top when it's not actually open source. Just some library they are using is open source and it's just a fork of sqlite
this guy is discovering that the water is wet... do we really need these kind of articles to state the obvious? apparently yes. There is CDN and and there is data-centers. Edge computing is a wicked concept. but apparently people need to spend millions before realize.
I lived there for a couple of years and got dirty looks for pronouncing it as Pierre my first month. I liked living in Sioux Falls more, outside of the scars of being a Cobol dev the first year I was there and then Java, so not much better.
No mention of the elephant in the room: who wants to use an Sqlite based database in a world where everyone is used to the capabilities of eg. Postgres.
ok real talk, i wasn't paid to read this. yeah i am sponsored on various things for the overall company behind it, but this was a "non sponsored" content. reasonably speaking, do i put ad or not on it? i genuinely don't know
@@ThePrimeTimeagen you should clarify your relationship with turso at the beginning of the video similar to linustechtips does with his investment in framework even if the video is not directly sponsored. you obfuscated this relationship by saying "this is my favorite database", "he is a very nice guy" and "turso engineers are very smart people" without disclosing the sponsorship. you also did it again in the pinned message e.g. you did not reach out to "the author", but to your sponsor. this is a problem, because i watch your videos to learn about new tech. these statements make it look like this is the best thing ever and i should check out libsql and turso. but not disclosing the relationship undermines your credibility a lot for me, because i have now to double check every project you talk about.
i reached out to the author and he said he used only free tiers for this test. Hence the reason there is no planetscale replication as its not available in the free world.
When people mentioned the 't' in Toronto, you should have started saying "Oronto" instead xD
What about supabase replication? Supabase got destroyed here but it didn't look like they enabled replication.
@Cookie note that all setups were provided by the database vendors themselves! (author here).
This is not a benchmark where we set up other vendors, this is organized by Vercel and each vendor set up their own systems.
18:11 I looked at the docs and Turso is essentially a read only replica as well. The difference is for writes, you still use the edge as an endpoint but it sends the write over to the primary.
ok, that is pretty neet. but good to know.
Not sure how planet scale works but I guess the benefit or turso is that you can just write to the replica and call it a day. You can offload the slow request to the primary to turso. I’d be interested in how they keep their system fault tolerant so that they don’t drop writes if a node fails
For optimistic updates I bet this would still slap (ie experiences like Linear)
Brazil mentioned! Yeahhh, come to Brazil Prime, we will give you some Pão de Queijo and Coxinha with caldo de cana. During the night time you will drink Cachaça💚💛💙
This is aewsome! As a software engenieer based in Brazil, we plan our applications to make as few requests as possible to anything in America and in many ocasions the latency is a deal breaker. Truely life changer lol
+ Turso website is the fastest one i visited in a looooong time
@Tunema hey - author of the blog here. I was actually born and raised in Brazil, although by now my Portuguese sucks a bit! Glad to see you liked the website
coé parça
@@crazycoder9356 doidera meu brod
Eager to see Prime turn back into his primary for of 'cat'
(for context, the title at the time of watching is "Benchmarking Databases on Vercel | Prime Recats", in case Prime fixes it later)
The idea of having read replicas everywhere sounds great in theory, but replication lag can be a killer. Especially for those cases where you require your data be fresh everywhere and have to enforce sync writes (and sync replication).
If your data can be a bit stale though, and use async replication, the more replicas the merrier.
If the data need to be fresh then you just need to read directly on the write connection. Does it work in this case?
I question when this actually is important. Stock tickers? Your avg CRUD SaaS probably won't care even for up to a minute lag. >5min and we have problems tho for sure.
Turso is interesting, they use raft consensus as far as I can tell to synchronize the various read state replicas across edges. On one of my previous projects, we combined raft with CRDTs to build a time series database and let me tell you, I wish this thing was not proprietary because I would absolutely be pushing it as the best distributed time series database out there. It just makes sense to do time series this way since you rarely are deleting or mutating data in a time series table. Anyway, I also wish that turso was open source.
you can always ask Glauber on twitter. he is pretty open about a lot of the decisions they make
@@ThePrimeTimeagen hmm, might be worth getting an actual Twitter account one of these days...
NGL prime your literally the Dr Dis of programmming,love that i came across your channel had fun learning and getting enterained these past few days
I came here to find hot databases in my area, not disappointed
Your pronunciation of Rio de Janeiro is suprisingly good
I'm super interested in Planetscale, though they don't support foreign keys and it just freaks me out lol. I've never used a db without fks.
Even Oracle didn't support FKs until v8.x. From Oracle v4 through v7 it was a royal handicap not having them.
Watching this from Australia in 240p because of how bad the internet is
12:20 @ThePrimeTime, could you pls make a video about what defines a really good engineer and how can achieve that level?
7:39 pure?
Great experience working with Turso so far, definitely suggest trying it out, def tried this out b/c of Prime.
Anyone know of write speed graphs please?! Wouldn’t Turso writes be slower?!
The benchmark should have been done with Planetscale's read replication turned on for it to be a fair comparison. I'd be very interested to see how that would affect the results.
yeah, the 1v1 results show a big difference as is
but the read replication is a pay for feature only. i reached out to the author and he said he used only free tiers for this test.
Author here! The tool is provided by Vercel, not us, and each database contributed their own setups - including Planetscale. That said, this is not really a benchmark against Planestcale - they actually did fantastic !
You are right that with read replicas planetscale (and others) can achieve this, but there is a reason they - and others - didn't add replicas in this test. It's quite expensive!
The advantage of Turso, that I tried to touch on the article, is how easy and cheap (the free tier alone already has 3 locations) is to set up replication.
Great comment, though! ❤
marketing benchmarks always have some amount of bs in them.
Remember, the name…is the Pruh-mEE-gen!
This is interesting because of the freemium perspective, but it doesn't bring fundamental shifts. I'm more interested in seeing how Cloudflare would do with their Distributed SQLite approach if they really get to make read/write replicas.
At this moment, No-SQL alternatives like FaunaDB or CoakroachDB seems to be the ones actually breaking the molds with respect of edge data.
Is the added complexity of the edge and distributed systems worth the small performance gains? Maybe? Maybe not?
or keep everything local, like in an android app
Why should I use Edge to access my database 😭
7:18 You could go 1 step further and pronounce Detroit as De-true-ah
Like Vuk Karadžić said: "Write as you speak and read as it is written."
Wonder how this compares to Cockroach DBs offerings ?
Why shouldn't you deploy at the edge ? Just read the pretty inconvenient cons on Vercel's own page ( + ignoring the fact that you have to write JS on the backend 😅)
Yeah the JavaScript thing kind of sucks
Doesnt vercel edge use CF workers which supports rust?
Prime recats
Why don't you like C++? I know Rust is great and doesn't have hidden execution paths, but at least in C++, you can recover from Out of Memory errors. (Just to note, I really like Rust.)
C++‘s share Pointers are thread safe by default meaning they have extra over head on construction because they are atomic. Rust has Rc and Arc for specific use case. Rust io and string manipulation is leagues ahead of C++. The compiler, build tools, packages and ecosystem are better. CMake is horrible, i build recursive globing smart CMake scripts with for loops macros function… it’s horrible. Cargo is a dream to work with. Again the dependency management is just better. And Rust is just plain more fun to write in
Other than terrible ecosystem/toolchain, C++ is extremely hard to learn and extremely easy to cause some serious issues in without even realizing it. It's super feature-bloated and those features are not easily discoverable, a lot of the more modern changes are good but at this point it's almost impossible to know what's best practice, what you should do and what you should avoid, reading someone else's code can be hell even if you're already very proficient at it. There's less and less junior C++ jobs nowadays, because juniors are useless at writing C++, you're either very good at it or you're useless, or even potentially detrimental to the project if your mistakes go unnoticed. Since better alternatives exist nowadays, there's simply no reason to use it, and no reason for new devs to learn it.
won't that be very expensive? You literally have a database per edge + cross region costs are pretty high in the cloud so that will add another hidden cost just to get a couple of milleseconds.
I walked on a Pierre and watched some boats
Pretty misleading messages on their landung page. They put the text "Open source" at the top when it's not actually open source. Just some library they are using is open source and it's just a fork of sqlite
Canadians are just excited they got referenced at all
this guy is discovering that the water is wet... do we really need these kind of articles to state the obvious? apparently yes. There is CDN and and there is data-centers. Edge computing is a wicked concept. but apparently people need to spend millions before realize.
As a Helena,MT guy … I’m concerned
tiger beetle is real
it happens once every twice
Blazingly fast
I lived there for a couple of years and got dirty looks for pronouncing it as Pierre my first month. I liked living in Sioux Falls more, outside of the scars of being a Cobol dev the first year I was there and then Java, so not much better.
YOU ARE GOING TO BRAZIL
I want to see the graph of viewers anger after the pronunciation section! 😂😂😂😂
Prime Recats 🐱
Chile mentioned
RECATS
Hello from SANTIAGO DE CHILE
JAY SAHN
No mention of the elephant in the room: who wants to use an Sqlite based database in a world where everyone is used to the capabilities of eg. Postgres.
pee-air
you should disclose the sponsorship. this is basically an ad
ok real talk, i wasn't paid to read this. yeah i am sponsored on various things for the overall company behind it, but this was a "non sponsored" content.
reasonably speaking, do i put ad or not on it? i genuinely don't know
@@ThePrimeTimeagen you should clarify your relationship with turso at the beginning of the video similar to linustechtips does with his investment in framework even if the video is not directly sponsored.
you obfuscated this relationship by saying "this is my favorite database", "he is a very nice guy" and "turso engineers are very smart people" without disclosing the sponsorship. you also did it again in the pinned message e.g. you did not reach out to "the author", but to your sponsor.
this is a problem, because i watch your videos to learn about new tech. these statements make it look like this is the best thing ever and i should check out libsql and turso. but not disclosing the relationship undermines your credibility a lot for me, because i have now to double check every project you talk about.
brasil ? BRAsil ! BRASILLLLLLLLLLLLLLLLLL !!!
Recats
meow
redis is best
I always called it "pee air" ever since I was little (but I'm from WV and maybe we all have fukd accents)
that is the only way to pronounce it
shout louder XD
bruh
Bro looks and sounds like drdisrespect