[mapserver-users] Mapserver Performance Statistics

woodbri at swoodbridge.com woodbri at swoodbridge.com
Mon Sep 9 08:21:43 EDT 2002


Pradeep,

Much of the testing results and stats you are asking for will be 
highly related to the specific hardware setup the was tested and 
probably can not be translated to a different hardware configuration.

I did some stress testing a while back on a mapserver configuration 
using something called siege. The  most  recent  released  version  
of  siege  is available by anonymous ftp from ftp.armstrong.com in 
the directory pub/siege.

The thinks that determine mapservers ability to handle requests are 
the same things for any other hardware, like number of cpus, amount 
of memory, speed of disk, load balancing across multiple systems. 
What your disk cache hit ratio is.

Things like its a CGI app don't enter the picture because that is 
what mapserver IS, sure if it could be a module it might run a little 
faster, but it isn't right now.

But probably the BIGGEST performance problem I have seen repeatedly 
with mapserver is that people that have not work extensively with do 
not know how to optimize the map file, tileindexes, spatial indexes, 
etc to make mapserver fast.

I was able to get the system I ran siege on to render consistently at 
8-10 secs for out of cache queries and 2-3 secs for cache hit 
queries. Is this good? Well that was down from 45+ sec per render 
before I started working with is. But is this good? It was as good as 
the piece of hardware could do, but it was not a server class 
machine, it was a desktop system with very slow IDE drives, no raid, 
single cpu, 1 GHz processor, 1 GB of memory, but I was able to get 
similar performance from my 450 Mhz, 128 MB box because it was 
equally I/O bound.

If you want to collect stats on your CGI hits, write a script called 
something like php2 and place in it:

   time  -o /tmp/time-stats.log --append /path/to/php $*

and place it in your cgi-bin directory and have your webserver 
execute it instead of php, or do something similar for mapserv and 
then run siege and you can compare the siege stats along with the cpu 
and i/o performance numbers.

Hope this helps,
  -Steve

On 9 Sep 2002 at 17:06, Pradeep wrote:

> 
> Hello,
>  We are evaluating the performance of Mapserver to go ahead 
> with plans of developing a product based on the MapServer core, and we
> are looking for statistics on the performance issues listed such as
> The No of requests per second, the success rate, the HTTP errors, the
> amount of processor overhead, the memory bloat...etc etc..
> 
> Can some one be kind enough to shed more light on any previous work
> already done on any performance issues of MapServer.... I would be
> glad to see figures related to performance, and a comparitive study of
> MapServer vis-a-vis any other similar product/tool available in the
> market.
> 
> PS: One moot question I would like to ask the folks all around is
> "What are the pros and cons of MapServer being a CGI Exe, is it an
> Ideal choice for a Web Application, as we know generally that all CGI
> apps have an inherent issue of getting loaded in to the memory for
> each request".
> 
> Regards,
> Pradeep.





More information about the mapserver-users mailing list