Hacker News Reader

Comment by Osiris | Hacker News Reader

I tried out GWAN on a VPS a while back. There's no configuration file, you setup directories named in certain ways to tell the webserver what IP/port to bind to.

I cannot comment on how much fast it is because it's been years since I ran benchmarks against it. Even supposing the claims are true, currently it's very limited in what languages it supports. It's great at static content, but dynamic content must be written in C or use a C library. In other words, PHP, Python, and Ruby support are pretty much non-existent (though someone has tried to get PHP FastCGI to work with it).

The author is also fairly arrogant. Earlier versions of the website claimed his code was "perfect" and that he wrote it because no else seemed competent enough to do it right (I'm trying to find the quote).

Regarding PHP: > And without something like G-WAN you can forget about PHP in the years to come because it will not survive the parallelism's requirements.

View on HN

you setup directories named in certain ways

I was intrigued by this and had a look. One level of the hierarchy requires directories either named with a starting # or $. !!! Possibly two of the worst leading characters to use if there's ever a chance one of these file names is going to hit a shell script. Or the command line. And since the point seems to be you can adjust the config by pushing things around with mv, seems like that's going to happen a lot.

Rule number one when putting your metadata in a filename is to not use shell metacharacters.

> Even supposing the claims are true, currently it's very limited in what languages it supports. It's great at static content, but dynamic content must be written in C or use a C library.

That is no longer accurate. G-WAN now supports C, Java, C++, D, and Objective C out of the box. [1] I am the someone who is implementing FastCGI support.

[1]http://gwan.ch/developers

I'm not trying to be disrespectful because I'm sure you've put in a lot of work on your part of the project, but little to nothing on that page makes any sort of sense. Where is this terrifying efficiency hit from "the frenetic wave of abstraction layers" that the author of that page is claiming? Does the author even realize that he sounds like a huckster and his claims about other projects have only a tenuous relationship to reality? And why would I trust the judgment of developers who think that writing web applications in C is a perfectly cromulent idea?

Second reaction: am I actually supposed to take seriously something that looks like that Java sample? 'Cause, er, I know a teeeeeensy bit about Java, and my first reaction was "what kind of moron would write Java like that?". Java isn't C, and this guy's bizarre worship of "programming to the metal" (it's a web architecture for god's sake!) makes no sense there.

From my preliminary read of that page and the linked loan.java, it looks like the java is actually being processed into C code and then complied by the server. Can anyone confirm if this is how the system is architected to "support" java?

It's honestly hard to tell, because behind the SEO-dross and random attacks on established, already-high-performance tools, there aren't many details at all. At first (I hadn't looked at the linked file) I immediately said "you have to be kidding me", but...then I looked at the Java file and I think you might be more on-target than I'm comfortable with.

It does not look like it's running in a normal JRE, which makes me really skeptical of the sibling post claiming that "you can write Java however you want". (If it was running in a normal JRE, a ton of the stuff in loan.java makes little sense...)

The FAQ says it depends on either OpenJDK or the official Oracle Java, so it doesn't sound like it's a special Java-to-C compiler, unless it's something like a runtime JVM bytecode-to-C compiler.

That could be the case--I'm not going to dig deeply into it because honestly I like having my brain still work, but it could be that the insanity in loan.java is just because of the crappy C API being wedged into Java.

I'm not exactly sure how his claims of instant refreshing "scripts" works if it's a conventional JDK, though. I mean, that's just not possible, even with hot reloading (as Play Framework developers have learned to their disappointment).

It wouldn't surprise me if it has something like GCJ creating something like a .so that it loads on the fly. That is just a thought though, i haven't inspected it particularly closely.

Based on the graphs, the contents, and the over-the-top design, I was wondering if this was a joke or something until seeing this thread. So I'd agree that alarm bells may be warranted.

On the other hand, I have written a few web applications in C. It's not that bad! :) The nice thing about it is that if you link statically, write CGI programs, and ignore memory management (or, rather, malloc all you care to and delegate garbage collection to the OS when your process goes away), you can get something that runs quickly with minimal hassle on a small device. I don't recommend it unless you have some bizarre constraints, but it's really not that bad! That approach doesn't seem possible with this server, sadly.

Sure, you can do it, but CGI is pretty much dead as a "fire up a process, do something" method of serving web content. If the request processor isn't also your HTTP server, you'd generally start up a FCGI daemon that handles requests.

At that point the amount I trust 99.9% of C and C++ programmers to not screw the pooch (between dealing with HTTP requests, database interaction, string templating...) trends very close to zero. What you describe might be workable in the small, but at that point you might as well fling up Apache and mod_php. It'll probably be faster.

If CGI is dead, I had not noticed. :) I'll admit that I don't write CGI programs often, but I do still write them.

I think, though, that the database interacting, templating, etc., are things that you will very nearly always do in PHP because they're there. File-backed storage (if any storage is even needed) is more than adequate for most cases where one might write a CGI program in C. The last time I had occasion to write something from scratch and target CGI, I was generating graphs. No templating, no database, very fast.

The major pain of templating lies in C's string handling, though, and if you aren't worrying about freeing memory, there's not much pain. As far as HTTP goes, if the protocol was designed to be parsed in C, it's not often difficult to parse in C. But for all of those things, there's a library.

I can't comment specifically about whether or not mod_php would be faster (since I used PHP only briefly and several years ago) but I suspect very strongly that the overhead of firing up a small, statically linked C program beats it.

As I said previously, I don't think it's usually a good idea, but it's a more than viable tool and nice to keep in the box.

I did note in the comment about PHP support. I did notice the website mentioned C++/ObjC, but I just lumped those in as C variants. The Java integration looks interesting.

The author in a forum post did say that he doesn't think that FCGI is the right way to go and suggested asking Zend to build a C library for GWAN that could run PHP code [1]. I find that unlikely unless GWAN really starts getting a large following.

[1] http://forum.gwan.com/index.php?p=/discussion/comment/3801/#...

PHP support is important for backwards compatibility. This is one reason why I am implementing FastCGI.

See [1] for reasons why PHP+FastCGI is a poor choice in 2012.

[1]http://gwan.ch/en_fastcgi.html

Actually, on the developers page you linked it also lists JS, Lua, Go and Python as languages for GWAN handlers. I assume that means that those languages are supported via separate processes that are called out by the GWAN process since the page breaks those languages out separately from 'natively' supported C/C++/Obj-C, D and Java.

I wrote the Lua, Go, and Python examples handlers.

JavaScript, Lua, and Python can be linked directly with the server for performance you would not get via separate processes.

I had more difficulty with Go, which you can read about on the forum.[1]

They all work, but are just proofs of concept at the moment. G-WAN makes it easy to link with any C library, so it is not hard to use languages that expose a C API such as those listed. It is just a matter of writing a more complete handler.

[1]http://forum.gwan.com/index.php?p=/discussion/463/writing-g-...

Reply on news.ycombinator.com