On 11/13/00 7:35 AM, Pierre Sahores wrote:
> LiangTyan Fui a �crit :
>>
>> On 11/12/00 7:28 PM, Pierre Sahores wrote:
>>
>>> andu a �crit :
>>>>
>>>>> Yes, I've been using the AppleEvent handle on Mac for CGI some time ago.
>>>>> But
>>>>> the major concern here is the complexity of the CGI. Suppose I have a
>>>>> chunk
>>>>> of code that would takes a long time to execute (10 seconds?), the second
>>>>> query from the AppleEvent will have to wait.
>>>
>>> You mc code need optimisation. I would never serve something on the web in
>>> using
>>> a piece of code unable to reply in more than 50/60 ticks.
>>
>> Even if I can cut each query down to 30 ticks, I'll have problem when I have
>> 5 to 10 hits per seconds - and yes, it happens.
>
> Congratulations if your server runs with such an hits average. For 10
> hits/second * 3600 * 24 = 864000 hits/day, both macos 8/9 and winnt are going
> to be out of the range and you will need to use an unix server. If the hits
> average is < 150000/day, i just know that MacOS 8.1 and up + WebStar 3.02
> (reply timeout set to 900 seconds) + HC/PopToFront or MC will run fine, even
> if some requests are going time-to-time to a slowest service (like
> Lasso/FileMaker) running on the same G3 266 Mhz/292 Mo server. If you want to
> get the same average from a winNT monoprocessor server without crashing it day
> after day, don't try to use any web application server solution (Websphere,
> Oracle, Metacard, etc..) nor any cgi config and just run little ASP/SQLServer
> solutions.
Most of the hits were not coming from the public. The CGI serve webmail for
11 thousands student in a university, based on a 10BaseT Ethernet, so you
can imagine the traffic.
>> On an on going development project, I just don't have luxury to think
>> "optimised" over "functional".
>
> I agree and when i have to write 280000 chars of mc code in a month, i'm happy
> to know that the stuff will be web served by a Linux box ;-)
I just started intensively testing MetaCard socket on Linux, maybe I'll port
the webmail CGI to Linux based on MetaCard socket, integrated with MCHttpd
if possible.
>>>>> In most case this is ok if I plan carefully, but when I start deploying
>>>>> complex solution this way, I have to always ensure:
>>>>> - each query has to be short
>>>
>>> Use only "POST" procedures and boost the Ram over more than 256 Mo.
>>
>> Can you elaborate how "POST" is better then "GET"?
>
> 1.- Because each post request need to be recognised by the web application or
> cgi script by its name/value key, the server will not answer anything to an
> unrecognised key (server more secure againt attacks where it's always easiest
> to crashe a server in sending it a non autorised get request).
>
> 2.- Because the length of a post request is only limited by the ram avalaible
> on the server.
If this is the case, there is no speed different issue I presumed?
<snip>
Regards,
LiangTyan Fui
Archives: http://www.mail-archive.com/[email protected]/
Info: http://www.xworlds.com/metacard/mailinglist.htm
Please send bug reports to <[EMAIL PROTECTED]>, not this list.
