[Openstreetmap-dev] Re: Ideas to lower server resources

[M Josenhans]

Hi,

 > Anyone with profiling experience around?
I have no ruby profiling experience, however automatic collection of 
profiling data during normal operation in conjunction with  automatic 
generation of statistical performance graphs for different time 
intervals (hourly, daily, weekly, last week, etc.) helped in similar 
cases to achieve significant performance improvements over the time.

Collecting this statistical data and having it automatically published 
on a web page, helps to determine changes (hw, sw, etc.) with negative 
performance impact and allows immediate correction those.

Michael

Immanuel Scholz wrote:
> Hi,
> 
> moved to dev.
> 
> 
>>> If donations don't get enought money, we need to change the software so
>>> it use not that much resources until donations are enough. ;-)
>> Please do!  What changes would you make, and can you make the case
>> that those features aren't necessary?
> 
> A very good approach that greatly worked in the past to those things is
> distribute the whole effort. Networks like Gnutella exist, because people
> want the thing they provide and because all people are willing to
> contribute to the (MASSIVE!) need of ressources as bandwidth and CPU
> power.
> 
> Well, I think this is a very long-term goal and some kind of a "last
> effort", since building good heavy distributed systems are much much more
> complex than building good centralized servers.
> If the overhead we need in programming power only for maintain the
> distribution is larger than the resources we need to maintain a server,
> this is no solution.
> 
> 
> Caches. We need good caches! As example for the own people at home. If
> people can just download an exe (or aptitude a package) and have a local
> cache of the tile server, I think this can improve performance by several
> magitudes.
> However, I did not profile this, so maybe I am wrong here, but still it
> sounds like a good mid-term solution to me.
> 
> 
> Last time I looked at the ruby code, I was able to speed up the map.rb -
> requests by more than 30% with some simple changes (considering the very
> basic and possible wrong measurements I made ;( ). I think there is still
> great potential to just optimize the server to be not that resource
> hungry. This is the short-term solution, we have ;)
> 
> 
> I think better profiling is an important goal. Especally when we are
> planning to change something (other hardware, other database software,
> other server programming language, other protocol API etc.) we need the
> posibility to compare the results. I can only speak for myself, but this
> is the main reason, I am currently not interested in doing performance
> things: I am unsure whether this has any impact or may just worse the
> whole thing.
> Anyone with profiling experience around?
> 
> 
> Ciao, Imi.