|
|
@@ -1,11 +1,82 @@
|
|
|
# Lwan
|
|
|
|
|
|
-This is the configuration files to benchmark the [Lwan](http://lwan.ws)
|
|
|
+This is the configuration files to benchmark the [Lwan](http://lwan.ws/?src=twfb)
|
|
|
web server/framework.
|
|
|
|
|
|
+## FAQ
|
|
|
+
|
|
|
+### Where's the source code for the benchmark?
|
|
|
+
|
|
|
+Source code for the benchmark itself is currently living in [in Lwan's
|
|
|
+repository](https://github.com/lpereira/lwan/tree/master/techempower) for
|
|
|
+practical reasons, but will move to [this repository right after Round 10 is
|
|
|
+complete](https://github.com/TechEmpower/FrameworkBenchmarks/issues/1372).
|
|
|
+
|
|
|
+In any case, the script to install the framework in the test machines [pulls
|
|
|
+a specific snapshot of the Lwan source
|
|
|
+code](https://github.com/lpereira/lwan/tree/49607addb31879e2aa2b701317773674662315aa),
|
|
|
+including the benchmark code; and, due to the nature of Git, it's not
|
|
|
+possible to alter that code without changing which commit hash to checkout
|
|
|
+in this repository.
|
|
|
+
|
|
|
+### The JSON test RPS value seems ridiculous; is Lwan cheating?
|
|
|
+
|
|
|
+Come on. There's no optimization without cheating. There's a somewhat
|
|
|
+[detailed blog post about
|
|
|
+this](http://tia.mat.br/blog/html/2014/10/06/life_of_a_http_request.html?src=twfb).
|
|
|
+However, that level of cheating is considered by Lwan's author to be fair
|
|
|
+game, since it's not related to the benchmark code itself.
|
|
|
+
|
|
|
+In fact, the serialization library has been picked solely because it was
|
|
|
+available in the [CCAN](http://ccodearchive.net/). It's written by [Joseph
|
|
|
+A. Adams](http://ccodearchive.net/info/json.html), and has been slightly
|
|
|
+tweaked to fix some compiler warnings and a few other minor changes. No
|
|
|
+major optimization or trickery has been made there: it's essentially the
|
|
|
+same code. Props to him for a clean code that performs well.
|
|
|
+
|
|
|
+It's all open source anyway, feel free to go ahead and inspect the code.
|
|
|
+
|
|
|
+### The database results aren't that good. Why?
|
|
|
+
|
|
|
+The database tests are included for completeness sake. The database
|
|
|
+abstraction layer (which works with SQLite and MySQL) was written
|
|
|
+specifically for this benchmark, and is completely synchronous: while
|
|
|
+one query is being performed, all other clients wait.
|
|
|
+
|
|
|
+So the uninspiring results are expected. Database connectivity is
|
|
|
+important and will eventually be revisited.
|
|
|
+
|
|
|
+### I've seen the preliminary results and Lwan didn't fare well in the plaintext benchmark. Is there a reason why?
|
|
|
+
|
|
|
+These benchmarks are performed using pipelined requests. For practical
|
|
|
+reasons (almost no client supports them, at least by default), this was not
|
|
|
+implemented in Lwan: the server was dropping the request buffer and waiting
|
|
|
+for the next request to come in, while `wrk` was waiting for responses to
|
|
|
+come out, eventually timeouting and sending more pipelined requests, only to
|
|
|
+repeat, until the test actually ended.
|
|
|
+
|
|
|
+Pipelining has been implemented in a hurry to get included in Round 10.
|
|
|
+
|
|
|
+### The results are pretty good, can I use this in my next web app?
|
|
|
+
|
|
|
+Writing web apps in C? Only if you're Crazy. (The uppercase C is not a typo.)
|
|
|
+
|
|
|
+Seriously, Lwan is just a toy. It's an experiment. Most things you'll need
|
|
|
+from real frameworks won't be available. You'll have a hard time trying to do
|
|
|
+the simplest things. There are far more important things to look for while
|
|
|
+picking a framework, and performance, while important, is not the most
|
|
|
+important factor. However, if you're feeling adventurous, know that Lwan is
|
|
|
+free software, and is very open to pull requests.
|
|
|
+
|
|
|
+### I have other questions, is there a FAQ somewhere else?
|
|
|
+
|
|
|
+Sure there is. [See Lwan's web page](http://lwan.ws/?src=twfb) or contact the author
|
|
|
+directly.
|
|
|
+
|
|
|
## Requirements
|
|
|
|
|
|
-GCC, SQLite 3.7, Linux 3.0, optionally jemalloc or tcmalloc.
|
|
|
+GCC, SQLite 3.7, MySQL/MariaDB client libraries, Lua 5.1/LuaJit 2.0, Linux 3.0,
|
|
|
+optionally jemalloc or tcmalloc.
|
|
|
|
|
|
## Tests available
|
|
|
|
|
|
@@ -29,7 +100,6 @@ URL: /fortunes
|
|
|
|
|
|
URL: /plaintext
|
|
|
|
|
|
-
|
|
|
## Contact
|
|
|
|
|
|
-Leandro Pereira <[email protected]>
|
|
|
+[Leandro Pereira](http://tia.mat.br/?src=twfb) <[[email protected]](mailto:[email protected])>
|
Hamilton Turner