talksik/uWebSockets
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

New chart

Browse Source
This commit is contained in:
Alex Hultman
2018-10-22 17:41:00 +02:00
committed by Alex Hultman
parent ee43523438
commit 06f7b04810
2 changed files with 6 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
benchmarks/README.md
+5 -3
View File
@@ -6,14 +6,16 @@ The world needs more like you. I could talk about how stupid and gullible 95% of
* Never accept random *numbers* which has no *point of reference* to similar software projects. The number "1 million" might sound fancy but without *any* kind of reference to *other* projects there's no knowing the worth of such a number. Comparative benchmarks are the only ones worth taking in. Think for yourself - how can the author claim "fast" if he/she has no benchmarks of what *others* are doing?
* Most people who do benchmarks are complete monrons and do not know their ass from their head. Always make sure to *normalize* performance by spent CPU-time. That is, do NOT compare performance of software demanding 100% CPU-time on 57 CPU cores, with software running on 5% CPU-time on 1 CPU core. You would be stunned if you knew how few people actually care to even check this before posting their results to their literally millions of gullible followers.
### The bystander effect
![](../misc/bigshot_lineup.png)
### The bystander effect (todo: shorten this)
Having many users and many "stars" does not constitute a confirmation of any such *claims*. In fact, the opposite is more often true than not. How can this be true? Well for starters, actual performance often come with restrictions and more cumbersome APIs and it turns out most people are not really interested in doing more work than what they absolutely have to. Programmers want to *believe* their overly abstract garbage collected functional one liners to be "more efficient than machine code". They want to have the cake, and a cake cloning device too.
Also, there's such a fenomenom where we humans are *less* likely to help someone in need of help when there are others present. We like to think to ourselves that, with this many people around, there *surely* has to be someone already helping that poor fellow. The thought of someone dying of blood loss and nobody helping is so absurd to people that they immediately refuse to even consider that case.
The same goes with software projects with massive amounts of popularity. People think that it *cannot possilbly* be true that such a massively popular project is *completely* trash in regards to performance! *Surely* it must be true.
The same goes with software projects with massive amounts of popularity. People think that it *cannot possilbly* be true that such a massively popular project is *completely* trash in regards to performance! *Surely* it cannot be *this* bad!?
The perfect example is Socket.IO - a project downloaded millions of times every single day. Books have been written, people use it everywhere. It *claims* using superlatives to be *the fastest in the universe and all other universes and across all space and time* yet delivers such a dreadful performance you would need close to 200 mondern computers running egg fryingly hot just to saturate a mesely 1 Gbit network with small messages. I consider myself an open minded person but even I have problems wrapping my mind around such a massive load of insanity.
The perfect example is Socket.IO - a project downloaded millions of times every single day. Books have been written, people use it everywhere. It *claims* using superlatives to be *the fastest in the universe and all other universes and across all space and time* yet delivers such a dreadful performance you would need close to 200 modern computers running egg fryingly hot just to saturate a mesely 1 Gbit network with small messages. I consider myself an open minded person but even I have problems wrapping my mind around such a massive load of insanity.
### Enough about psychology
The performance of µWebSockets is in most cases what you would expect from an efficient server software. In some cases, especially small messages, the difference in performance can be dramatical (10 - 15x). Because of the nature of WebSocket messages (they are mostly very small at sub 1kb), this plays a major role in performance.