Conference gyro::internet_server

    
    
    Hello again,
    
    As I couldn't find anything wrong, we decided to reboot the system at
    noon and the reboot cleared the problem. Could it be there is a
    potential problem here ? Need more info ? Tell me.
    
    Regards,
    
           Guy

In the past we have been telling customers to disable caching
on the server. We haven't received any complaints that the
described fix wasn't working. If you look at the netscape.server
newsgroup, they also tell you to do the same. The first attachment
is from the netscape.server newsgroup, the second one if from Netscape.

Hope this helps,

Shari
--------------------------------------------------------------

You need to add the following line right after the Init section in the obj.conf
of the port you are having troubles with.

Before:

Init format.access="%Ses->client.ip% - %Req->vars.auth-user% [%SYSDATE%] 
\"%Req-
>reqpb.clf-request%\" %Req->srvhdrs.clf-status% %Req->srvhdrs.content-length%" 
f
n="flex-init" access="/usr/internet/ns-home/https-IAS-Public/logs/access"
Init fn=load-types mime-types=mime.types

After:
Init format.access="%Ses->client.ip% - %Req->vars.auth-user% [%SYSDATE%] 
\"%Req-
>reqpb.clf-request%\" %Req->srvhdrs.clf-status% %Req->srvhdrs.content-length%" 
f
n="flex-init" access="/usr/internet/ns-home/https-IAS-Public/logs/access"
Init fn=load-types mime-types=mime.types
Init fn="cache-init" disable="true"



Attached is a response from a Netscape engineer.
-------------------------------------------------
Solution:
=========
> 
> Yup.  These are related to server side caching.  We have not had a
> whole lot of success with disabling server-side cache on Digital:
> 
> Init fn="cache-init" disable
> 
> does not seem to work on digital or AIX.  The server does not complain
> about the syntax, yet it has been known to rewrite the directive to:
> 
> Init fn="cache-init" 2=disable
> 
> on later save / apply when it is modifying obj.conf.  It's as if the
> admin server expects all the parameters to be name / value pairs.  We
> have also tried:
> 
> Init fn="cache-init" disable=true
> 
> without success.  What appears to be happening is that the server is
> allocating more RAM for cache than the OS will allow.  I have had
> success by tuning enterprise and minimizing the amount of caching done
> by the server like so:
> 
> Init fn="cache-init" cache-size="32" mmap-max="1024"
> 
> The less the server caches, the less likely it is to run into low
> memory in the Unified Buffer Cache in digital's virtual memory
> scheme.  A setting like the one above is generally successful at
> eliminating this problem.
> 
> There are basically three tunable parameters in digital unix that may
> help alleviate the problem without having to modify the cache
> settings.  I've had varying degrees of success with this:
> 
> vm:ubc-maxpercent -  This is the percentage of blocks in the UBC that
> the file system is allowed to occupy.  It defaults to 100 - so if the
> file system does occupy 100% of the UBC there is no room for the VM to
> cache.  Digital says 50 is a safe value and you can adjust accordingly
> from there based on performance.
> 
> vm:vm-mapentries -  This is a key one in relation to this problem.  It
> is the number of memory-mapped files available per process.  It
> defaults to 200.  Our cache defaults to 512.  If our server attempts
> to cache more than what this is set to you will see these errors.  I
> typically adjust this to 20000 :)
> 
> vm: vm-vpagemax - Sets the maximum number of pages allowed in a
> process's address space.   According to digital, this should always be
> set to a value equal to vm-maxvas/8192.
> 
> Two other digital parameters that are key to tuning for our server,
> but not really related to this problem are:
> 
> proc:max-proc-per-user - Key in a heavily multihomed environment.  How
> many processes is the web server allowed to run?  Defaults to 64
> 
> proc:max-threads-per-user - Very important to FT/ENT!!  Must be higher
> than the MaxThreads setting on the Netscape server.  If multi-homing
> or multi-process configuration, this must be greater than summ of all
> maxthreads.  Defaults to 256.
> 
> As to this particular customer, She opened a call with us yesterday on
> this issue.  I did not speak to her and the tech who did is not here
> so I cannot confirm or deny her tale of us pointing her to digital.
> However I read the case documentation and the tech who spoke with her
> gave her the syntax to disable cache and mentioned some of these
> tunable parameters.  The tech she spoke with is extremely good and
> I find itr hard to believe that he "had a snotty attitude about the
> problem."  Support is our business and we do not tolerate poor
> customer service.  I will follow up with him tomorrow, however, and
> make sure that we contact her again to see that her problem is
> completely resolved and that she is happy.
> 
> Hope this Helps,
> 
> Dan
> 
> P.S. if anyone has any insight as to why completely disabling server
> side cache on digital unix seems impossible, I'd love to hear it.  Our
> experience has been that it is a problem on AIX as well...
> 

    
    
    Shari, 
    
    Thank you for this, since monday that problem happens every day now,
    I'll try what you suggest and we will see ... I'll keep this note
    updated with success ;^) or unsuccess :-( results
    
    Rgds
    
           Guy