postgis cyclic performance hit

[Richard Duivenvoorde]

Hi Romolo,
sorry, you are right, I didn't provide very much info; Here is some extra:
My developer machine: WinXP, Microsoft IIS (!), postg*
Production: Winows Server 2003, IIS, postg* (client is running IIS, so 
final configuration should use IIS...)
Running mapserver as cgi in IIS.


I've done some more experimenting no:
- running the same posgresql-layers in APACHE on same machine is ok
- running a geotiff OR shape layer in IIS is ok

So it must have to do something with the 
IIS/mapserver/postgislayer-combination.

Looking at cpu-performance of memory, I do not see excess use of either 
of them.
But it looks like IIS must clean up something (cache, connections I 
don't know) after every 10 mapserver-cgi requests if the map contains 
postgis layers???
Only for the postgis layers, all other layers are fine??

I suppose there are more people who have deployed an 
mapserver/iis/postgis combi isn't it?

Richard

Romolo Manfredini wrote:
> The informations are very few, but such a behaviour may depends on several
> factors like:
> Few postgres connections
> then  if you are using apache
> bad configuration of apache (few childs)
> fast recycling of  servers (respaning child too often)
> are you using the cgi directly or some interface ?
> on my developer machine using winxp / apache / php / postgres / postgis /
> phpmapscript and persistent connections never had cyclic performance hits
> and response times are rather good with very large layers.
> BTW I had to patch tcpip.sys for half-open connection issues (1000
> connections allowed) as I had issues while testing with firefox configured
> for maximum performances.
> 
> ----- Original Message ----- 
> From: "Richard Duivenvoorde" <rdmailings at DUIF.NET>
> To: <MAPSERVER-USERS at LISTS.UMN.EDU>
> Sent: Thursday, April 19, 2007 5:00 PM
> Subject: [UMN_MAPSERVER-USERS] postgis cyclic performance hit
> 
> 
>> Hi List,
>> I'm running a developer machine windows, postgis/mapserver.
>>
>> It contains one small layer (50 points).
>>
>> If I run a ogc-mapserver request (in a browser) for an image of those
>> points, the performance behaves strange:
>> - about 10 responses ok: about a few hundred milliseconds
>> - then one VERY slow one: about 4 seconds
>> repeating that every 9-14 requests
>>
>> created gist index on geometry layer
>> created btree index on id column
>> gave postgresql some more memory
>>
>> still the same result.
>>
>> QUESTION:
>> - any people running mapserver/postgis on a windows machine and having
>> good response times
>> - somebody a clue to this cyclic 'bad response time' behaviour?
>>
>> Thanks
>>
>> Richard Duivenvoorde