World: r3wp

Petr;  DON'T  :)  Suggesting that people try and sneak REBOL past 
sysadmins is a really really really bad idea.  If they can't be convinced 
to try it, then change ISP, or try to convinvce them again.  Don't 
be making REBOL look all cracky.   That is the kind of maneuver that 
could sink the ship.  The rest of us like the ship, and we fly the 
flag of peace and truth, not the Jolly Roger.  :)

My uplink speed kinda (no, it pretty much completely) sucks but I 
offer free hosting to any rebol that wants it at peoplecards.ca. 
 I just ask for patience if a new service needs to be installed while 
I work out kinks and the user needs to know that it's home based 
with a not-so-speedy delivery pipe and I offer little in the way 
of frills; meaning it's sftp or ssh cli, not cPanel or other gui.

We fly the flag of peace and truth
.. hummed to the tune of "God bless America"

Don't forget the great big smiley...

Is that the tune that  sounds like "God Save The King?"'   If there's 
a sugar maple blight, 'The Maple Leaf Forever' will sound lame ... 
and they'll never see the Eastern Townships annexed by Vermont.  
'CGI'  does stand for 'Chat Gateway Interfarce' doesn't it ?

Anybody noticed CGI is back as a programming model?

Let me explian.... (the PITL in user.r reminded me to post this)....

First - do not virtualize OSes

1) Think multicore
2) think memory is cheap (2Gb per core)
3) Typically,  /Core  consumes 8 mb of memory

4) do not encap, use  amodule management system like my 'require 
or Ladislav's 'include

5) wrt 3 and 4: the OS starts using its disk cache etc. After a few 
hits these operations will be cheap

6) do all session mgt etc in a database => sales up as well, no state, 
share nothing

Now, what happens? The OS will start distributing the CGI processes 
over the multiple cores. Using the disk cache etc to speed loading 
times, enough memory per core on the processor. A 8Gb RAM quadcore 
should be able to run +- 1000 procs/sec (rough estimate). That's 
just one box, with that load it should be profitable. And as you 
obey rule 6, you can scale up and load balance pretty easily.

Winding down: Apache CGI with RSP is pretty good these days. If you 
combine:
- module management
- logging
- error handling
- session management
- database protocol (mysql://)
- CGI params handling

you can "just work".

not virtualising OS this days is imo a mistake, no? :-)

depends on what you gain

Are you sure OS distributes CGI processes to different Cores? Is 
e.g. Apache working that way?

If they are separate processes, the OS should balance over cores.

Gregg: really? I thought that the reason why R3 will use threads 
for tasking instead of tasks is, that OS can better balance threads? 
Anyway, those questions are for gurus, I can only wonder :-)

Threads are much lighter, but not as separate. I don't know details 
though. On a dual core with hyper-threading on, spawning multiple 
processes, I can see the load is spread.

petr, the processes are managed by the OS too. *obviously* the os 
will distribute processes among processors. (unless the os has no 
multiprocessor support, that is). distributing threads is more difficult 
(because of the shared memory), however all good threading implementations 
should do it, and if you programs the threads correctly you can get 
the performance boost.

Can anybody give me an exampkle setup + explanation of FastCGI + 
lighttpd with FastCGI; also on the Rebol side. I know Francois mentioned 
it was easy but I don't get how you can do adaptive spawning one 
the same listening port (e.g, 10 Rebol FastCGI processes listening 
on port 1026 or so)

I used fastcgi in the past, wich Apache, under linux. All modes worked 
fine IIRC. However, under windows, the implementation was crippled, 
and only external mode worked.

Also, there was problem, that the same client could be served by 
different process, so fastcgi guys implemented some kind of "afinity 
patch", kind of proxy, which then connected the same client always 
to the same process.

I don't think those 10 processes would listen on port 1026? In fact 
I don't know, how it is being done.

Hope you read http://www.rebol.com/docs/fastcgi.html

That REBOL doc should really answer your question. Simply put, in 
External mode, you do something like /path/to/script (that does not 
need to exist) and you direct it to certain, already running REBOL 
process. But - rebol has no tasking, so you have to handle accepting 
connections and multiplexing. It is like with Rubgy - unless you 
are finished, you are not available to other requests ...

Read that, but that is why the adaptive scaling with lighttpd is 
interesting if you put number of request/fcgi porcess on 1. Then 
the daemon scales for you

Maarten, I agree with your observation and you can even scale it 
more.


If you see a web-server as just a request dispatcher to CGIs and 
a fast-answering-machine for user-feedback (pages, forms etc.) you 
just need a small and "simple" one like Cheyenne. The CGIs can be 
distributed to different cores (through the OS) or even to different 
machines (via TCP/IP).

As dispatching requests is most likely much faster than processing 
a request, a single web-server should serve a lot of users and a 
bunch of machines do the processing.

This is the coarse grained multi-process approach.

With Cheyenne you can already have the main httpd process on one 
machine and task-handlers (RSP or whatever) on other machines 8)

I am close to autogenerating fastcgi processes, linked with Lighttpd 
configs and generating automagical includes that match the web server 
config  for encap/Pro.

I am also adding a routing scheme (dialect) so basically you redirect 
all traffic except static stuff to a fastcgi process, it comes in 
the router and then it checks on extension or path (REST) what to 
do.

The generating part will be asking you a few questions and the generate 
matchhing config files and binaries that you can copy to an empty 
Linux box with only lighttpd installed (or enginx)

Session management is database-backed, so scaling up is hiring VPS'es 
and putting pound in between

With the avereage memory use fo Rebol < 10Mb you can coompute how 
many users I can concurrently server for complex operations (100-200 
minimum), so every machine I hire can host 500 customers. That means 
that I should earn e0.50 customer to get  a decent margin (roughly).

earn = ask

I finally learned the trick: 1) create a good interface (but w eknew 
that) 2) use back-end technology that lowers the cost per user (i.e. 
enlarges the # users per box)

Maarten, this sounds very cool. So the goal is to have a scalable 
web-service framwork based on fastCGI and simple tools?

That's how I should have described it in one sentence.

But  yes: load-balancer -> webserver(*) -> FastCGI(*) -> MySQL

FastCGI is a rebol process with core enhancements, session mgt, RSP 
etc. I am also integrating autodoc from Gabriele so the files will 
be more "literate" and I have a module management system in place 
thathandles from interactive to encap.

I am using assembla.com from SVN and trac, the actual application 
I am building is for personal life management.

As a rebol process is only 10 Mb.... I can serve lots of users on 
cheap VPS's, load balance them, data backup in S3.  No others invited 
until I get things stable enough. eed to ge things going

Sounds very cool. Go for it!!

I don't understand it a bit. I can understand webserver, fastcgi, 
mysql part, but what is that load-balancer part? Client side?

or some special server part?

No, before the webserver, so you scale transparantly to multiple 
webservers (in my scenario each webserver effectively is the load 
balancer for X FastCGI rebol processes; it's how nginx and lighty 
work)

What are the best ways to protect source code from view in a cgi 
script?  When a script is made world-viewable, isn't that compromising 
the source pretty badly?

You want to set the file permissions for each script  globally executable, 
but not globally readable.

Also protect the cgi-bin folder from being read by the whole world...If 
you are using Apache, look at the IndexIgnore directive:

http://httpd.apache.org/docs/1.3/mod/mod_autoindex.html#indexignore

One of my clients updates his site via some tool, which always seem 
to add some space between the lines. After some time, the page is 
instead of 400 rows something like 13K rows - the size goes from 
cca 25KB to 100KB. So I wrote a cgi script, which reads index.html 
and removes blank lines. Everything is OK, when I run the script 
from the console. But when I run it via a browser as a CGI script 
call, it can't write the file. Dunno why - cgi-script is being run 
using -cs switch, I even put secure none in there, cgi-script has 
the same own, grp set as index.html, but I can't write it ....

Maybe I should try the trick to connect to the FTP account and upload 
it there, instead of rewriting? :-)

Check the permissions of the file to be written. Are you also sure 
the cgi script is being executed? Check its permissions too..