I'm developing a Rails 2.3, Ruby 1.9.1 webapplication that does quite a bunch of calculation before each request. For every request it has to calculate a graph with 300 nodes and ~1000 edges. The graph and all its nodes, edges and other objects are initialized for every request (~2000 objects) - actually they are cloned from an uncalculated cached graph using Marshal.load(Marshal.dump()).
Performance is quite an issue here. Right now the whole request takes in average 150ms. I then saw that during a request, p开发者_开发知识库arts of the calculation randomly take longer. Assuming, that this might be the GarbageCollector kicking in, I wrapped the request in GC.disable and GC.enable, so that the request waits with garbagecollecting until calculating and rendering have finished.
def query
GC.disable
calculate
respond_to do |format| format.html {render} end
GC.enable
end
The average request now takes about 100ms (50 ms less).
But I'm unsure if this is a good/stable solution, I assume there must be drawbacks doing that. Does anybody has experience with a similar problem or sees problems with the above code?
If it makes your app faster, then use it.
I would add an ensure
statement so that if any exception is raised you don't end up with disabled garbage collection.
def query
GC.disable
calculate
respond_to do |format| format.html {render} end
ensure
GC.enable
end
No real disadvantages, except that when re-enabled the GC will take longer to run.
There are a number of articles on the web on the tuning Ruby's GC. Take a look at them, and maybe you can remove those lines. =)
There's no way you can cache the results and redo the calcs on the background every few minutes?
It might looks stupid but in this case, I'll try to call a C function from your ROR. This solution is quite hardcore but it should give amazing performance results ;)
Your solution with ruby is not a long therm solution it's just a fix ...
精彩评论